blob: b9d7036c366596d5e1117069a63df752f82df107 [file] [log] [blame]
Ben Widawsky0260c422014-03-22 22:47:21 -07001/*
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
Chris Wilson8ef85612016-04-28 09:56:39 +010037#include <linux/io-mapping.h>
Joonas Lahtinenb42fe9c2016-11-11 12:43:54 +020038#include <linux/mm.h>
Chris Wilson84486612017-02-15 08:43:40 +000039#include <linux/pagevec.h>
Chris Wilson8ef85612016-04-28 09:56:39 +010040
Joonas Lahtinenb42fe9c2016-11-11 12:43:54 +020041#include "i915_gem_timeline.h"
Chris Wilsonb0decaf2016-08-04 07:52:44 +010042#include "i915_gem_request.h"
Chris Wilson84486612017-02-15 08:43:40 +000043#include "i915_selftest.h"
Chris Wilsonb0decaf2016-08-04 07:52:44 +010044
Matthew Auld2a9654b2017-10-06 23:18:16 +010045#define I915_GTT_PAGE_SIZE_4K BIT(12)
46#define I915_GTT_PAGE_SIZE_64K BIT(16)
47#define I915_GTT_PAGE_SIZE_2M BIT(21)
48
49#define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K
50#define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M
51
Chris Wilsonf51455d2017-01-10 14:47:34 +000052#define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
53
Chris Wilson49ef5292016-08-18 17:17:00 +010054#define I915_FENCE_REG_NONE -1
55#define I915_MAX_NUM_FENCES 32
56/* 32 fences + sign bit for FENCE_REG_NONE */
57#define I915_MAX_NUM_FENCE_BITS 6
58
Daniel Vetter4d884702014-08-06 15:04:47 +020059struct drm_i915_file_private;
Chris Wilson49ef5292016-08-18 17:17:00 +010060struct drm_i915_fence_reg;
Daniel Vetter4d884702014-08-06 15:04:47 +020061
Chris Wilson75c7b0b2017-02-15 08:43:57 +000062typedef u32 gen6_pte_t;
63typedef u64 gen8_pte_t;
64typedef u64 gen8_pde_t;
65typedef u64 gen8_ppgtt_pdpe_t;
66typedef u64 gen8_ppgtt_pml4e_t;
Ben Widawsky0260c422014-03-22 22:47:21 -070067
Joonas Lahtinen72e96d62016-03-30 16:57:10 +030068#define ggtt_total_entries(ggtt) ((ggtt)->base.total >> PAGE_SHIFT)
Ben Widawsky0260c422014-03-22 22:47:21 -070069
Ben Widawsky0260c422014-03-22 22:47:21 -070070/* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
71#define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0))
72#define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
73#define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
74#define GEN6_PTE_CACHE_LLC (2 << 1)
75#define GEN6_PTE_UNCACHED (1 << 1)
76#define GEN6_PTE_VALID (1 << 0)
77
Chris Wilsondd196742017-02-15 08:43:46 +000078#define I915_PTES(pte_len) ((unsigned int)(PAGE_SIZE / (pte_len)))
Michel Thierry07749ef2015-03-16 16:00:54 +000079#define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1)
80#define I915_PDES 512
81#define I915_PDE_MASK (I915_PDES - 1)
Ben Widawsky678d96f2015-03-16 16:00:56 +000082#define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT))
Michel Thierry07749ef2015-03-16 16:00:54 +000083
84#define GEN6_PTES I915_PTES(sizeof(gen6_pte_t))
85#define GEN6_PD_SIZE (I915_PDES * PAGE_SIZE)
Ben Widawsky0260c422014-03-22 22:47:21 -070086#define GEN6_PD_ALIGN (PAGE_SIZE * 16)
Ben Widawsky678d96f2015-03-16 16:00:56 +000087#define GEN6_PDE_SHIFT 22
Ben Widawsky0260c422014-03-22 22:47:21 -070088#define GEN6_PDE_VALID (1 << 0)
89
90#define GEN7_PTE_CACHE_L3_LLC (3 << 1)
91
92#define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2)
93#define BYT_PTE_WRITEABLE (1 << 1)
94
95/* Cacheability Control is a 4-bit value. The low three bits are stored in bits
96 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
97 */
98#define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \
99 (((bits) & 0x8) << (11 - 3)))
100#define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2)
101#define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3)
102#define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8)
103#define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb)
104#define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7)
105#define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6)
106#define HSW_PTE_UNCACHED (0)
107#define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0))
108#define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr)
109
Mika Kuoppalae7167762017-02-28 17:28:10 +0200110/* GEN8 32b style address is defined as a 3 level page table:
Ben Widawsky0260c422014-03-22 22:47:21 -0700111 * 31:30 | 29:21 | 20:12 | 11:0
112 * PDPE | PDE | PTE | offset
113 * The difference as compared to normal x86 3 level page table is the PDPEs are
114 * programmed via register.
Mika Kuoppalae7167762017-02-28 17:28:10 +0200115 */
116#define GEN8_3LVL_PDPES 4
117#define GEN8_PDE_SHIFT 21
118#define GEN8_PDE_MASK 0x1ff
119#define GEN8_PTE_SHIFT 12
120#define GEN8_PTE_MASK 0x1ff
121#define GEN8_PTES I915_PTES(sizeof(gen8_pte_t))
122
123/* GEN8 48b style address is defined as a 4 level page table:
Michel Thierry81ba8aef2015-08-03 09:52:01 +0100124 * 47:39 | 38:30 | 29:21 | 20:12 | 11:0
125 * PML4E | PDPE | PDE | PTE | offset
Ben Widawsky0260c422014-03-22 22:47:21 -0700126 */
Michel Thierry81ba8aef2015-08-03 09:52:01 +0100127#define GEN8_PML4ES_PER_PML4 512
128#define GEN8_PML4E_SHIFT 39
Michel Thierry762d9932015-07-30 11:05:29 +0100129#define GEN8_PML4E_MASK (GEN8_PML4ES_PER_PML4 - 1)
Ben Widawsky0260c422014-03-22 22:47:21 -0700130#define GEN8_PDPE_SHIFT 30
Michel Thierry81ba8aef2015-08-03 09:52:01 +0100131/* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
132 * tables */
133#define GEN8_PDPE_MASK 0x1ff
Ben Widawsky0260c422014-03-22 22:47:21 -0700134
Zhi Wangc095b972017-09-14 20:39:41 +0800135#define PPAT_UNCACHED (_PAGE_PWT | _PAGE_PCD)
136#define PPAT_CACHED_PDE 0 /* WB LLC */
137#define PPAT_CACHED _PAGE_PAT /* WB LLCeLLC */
138#define PPAT_DISPLAY_ELLC _PAGE_PCD /* WT eLLC */
Ben Widawsky0260c422014-03-22 22:47:21 -0700139
Ville Syrjäläee0ce472014-04-09 13:28:01 +0300140#define CHV_PPAT_SNOOP (1<<6)
Michal Wajdeczko1790625b2017-09-08 16:11:30 +0000141#define GEN8_PPAT_AGE(x) ((x)<<4)
Ben Widawsky0260c422014-03-22 22:47:21 -0700142#define GEN8_PPAT_LLCeLLC (3<<2)
143#define GEN8_PPAT_LLCELLC (2<<2)
144#define GEN8_PPAT_LLC (1<<2)
145#define GEN8_PPAT_WB (3<<0)
146#define GEN8_PPAT_WT (2<<0)
147#define GEN8_PPAT_WC (1<<0)
148#define GEN8_PPAT_UC (0<<0)
149#define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000150#define GEN8_PPAT(i, x) ((u64)(x) << ((i) * 8))
Ben Widawsky0260c422014-03-22 22:47:21 -0700151
Zhi Wang43958902017-09-14 20:39:40 +0800152#define GEN8_PPAT_GET_CA(x) ((x) & 3)
153#define GEN8_PPAT_GET_TC(x) ((x) & (3 << 2))
154#define GEN8_PPAT_GET_AGE(x) ((x) & (3 << 4))
155#define CHV_PPAT_GET_SNOOP(x) ((x) & (1 << 6))
156
Matthew Auld0a038522017-10-06 23:18:24 +0100157#define GEN8_PDE_PS_2M BIT(7)
158
Joonas Lahtinenb42fe9c2016-11-11 12:43:54 +0200159struct sg_table;
160
Tvrtko Ursulin50470bb2015-03-23 11:10:36 +0000161struct intel_rotation_info {
Chris Wilson7ff19c52017-01-14 00:28:21 +0000162 struct intel_rotation_plane_info {
Ville Syrjälä1663b9d2016-02-15 22:54:45 +0200163 /* tiles */
Ville Syrjälä6687c902015-09-15 13:16:41 +0300164 unsigned int width, height, stride, offset;
Ville Syrjälä1663b9d2016-02-15 22:54:45 +0200165 } plane[2];
Chris Wilson8d9046a2017-01-14 00:28:22 +0000166} __packed;
167
168static inline void assert_intel_rotation_info_is_packed(void)
169{
170 BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int));
171}
Tvrtko Ursulinfe14d5f2014-12-10 17:27:58 +0000172
Chris Wilson7ff19c52017-01-14 00:28:21 +0000173struct intel_partial_info {
174 u64 offset;
175 unsigned int size;
Chris Wilson8d9046a2017-01-14 00:28:22 +0000176} __packed;
177
178static inline void assert_intel_partial_info_is_packed(void)
179{
180 BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int));
181}
Chris Wilson7ff19c52017-01-14 00:28:21 +0000182
Chris Wilson992e4182017-01-14 00:28:23 +0000183enum i915_ggtt_view_type {
184 I915_GGTT_VIEW_NORMAL = 0,
185 I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info),
186 I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info),
187};
188
189static inline void assert_i915_ggtt_view_type_is_unique(void)
190{
191 /* As we encode the size of each branch inside the union into its type,
192 * we have to be careful that each branch has a unique size.
193 */
194 switch ((enum i915_ggtt_view_type)0) {
195 case I915_GGTT_VIEW_NORMAL:
196 case I915_GGTT_VIEW_PARTIAL:
197 case I915_GGTT_VIEW_ROTATED:
198 /* gcc complains if these are identical cases */
199 break;
200 }
201}
202
Tvrtko Ursulinfe14d5f2014-12-10 17:27:58 +0000203struct i915_ggtt_view {
204 enum i915_ggtt_view_type type;
Joonas Lahtinen8bd7ef12015-05-06 14:35:38 +0300205 union {
Chris Wilson992e4182017-01-14 00:28:23 +0000206 /* Members need to contain no holes/padding */
Chris Wilson7ff19c52017-01-14 00:28:21 +0000207 struct intel_partial_info partial;
Ville Syrjälä7723f47d2016-01-20 21:05:22 +0200208 struct intel_rotation_info rotated;
Chris Wilson8bab11932017-01-14 00:28:25 +0000209 };
Tvrtko Ursulinfe14d5f2014-12-10 17:27:58 +0000210};
211
Ben Widawsky0260c422014-03-22 22:47:21 -0700212enum i915_cache_level;
Tvrtko Ursulinfe14d5f2014-12-10 17:27:58 +0000213
Joonas Lahtinenb42fe9c2016-11-11 12:43:54 +0200214struct i915_vma;
Chris Wilsonbde13eb2016-08-15 10:49:07 +0100215
Mika Kuoppala44159dd2015-06-25 18:35:07 +0300216struct i915_page_dma {
Ben Widawskyd7b3de92015-02-24 16:22:34 +0000217 struct page *page;
Mika Kuoppala44159dd2015-06-25 18:35:07 +0300218 union {
219 dma_addr_t daddr;
220
221 /* For gen6/gen7 only. This is the offset in the GGTT
222 * where the page directory entries for PPGTT begin
223 */
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000224 u32 ggtt_offset;
Mika Kuoppala44159dd2015-06-25 18:35:07 +0300225 };
226};
227
Mika Kuoppala567047b2015-06-25 18:35:12 +0300228#define px_base(px) (&(px)->base)
229#define px_page(px) (px_base(px)->page)
230#define px_dma(px) (px_base(px)->daddr)
231
Mika Kuoppala44159dd2015-06-25 18:35:07 +0300232struct i915_page_table {
233 struct i915_page_dma base;
Chris Wilsondd196742017-02-15 08:43:46 +0000234 unsigned int used_ptes;
Ben Widawskyd7b3de92015-02-24 16:22:34 +0000235};
236
Michel Thierryec565b32015-04-08 12:13:23 +0100237struct i915_page_directory {
Mika Kuoppala44159dd2015-06-25 18:35:07 +0300238 struct i915_page_dma base;
Ben Widawsky7324cc02015-02-24 16:22:35 +0000239
Michel Thierryec565b32015-04-08 12:13:23 +0100240 struct i915_page_table *page_table[I915_PDES]; /* PDEs */
Chris Wilsonfe52e372017-02-15 08:43:47 +0000241 unsigned int used_pdes;
Ben Widawskyd7b3de92015-02-24 16:22:34 +0000242};
243
Michel Thierryec565b32015-04-08 12:13:23 +0100244struct i915_page_directory_pointer {
Michel Thierry6ac18502015-07-29 17:23:46 +0100245 struct i915_page_dma base;
Michel Thierry6ac18502015-07-29 17:23:46 +0100246 struct i915_page_directory **page_directory;
Chris Wilsone2b763c2017-02-15 08:43:48 +0000247 unsigned int used_pdpes;
Ben Widawskyd7b3de92015-02-24 16:22:34 +0000248};
249
Michel Thierry81ba8aef2015-08-03 09:52:01 +0100250struct i915_pml4 {
251 struct i915_page_dma base;
Michel Thierry81ba8aef2015-08-03 09:52:01 +0100252 struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
253};
254
Ben Widawsky0260c422014-03-22 22:47:21 -0700255struct i915_address_space {
256 struct drm_mm mm;
Chris Wilson80b204b2016-10-28 13:58:58 +0100257 struct i915_gem_timeline timeline;
Chris Wilson49d73912016-11-29 09:50:08 +0000258 struct drm_i915_private *i915;
Chris Wilson84486612017-02-15 08:43:40 +0000259 struct device *dma;
Chris Wilson2bfa9962016-08-04 07:52:25 +0100260 /* Every address space belongs to a struct file - except for the global
261 * GTT that is owned by the driver (and so @file is set to NULL). In
262 * principle, no information should leak from one context to another
263 * (or between files/processes etc) unless explicitly shared by the
264 * owner. Tracking the owner is important in order to free up per-file
265 * objects along with the file, to aide resource tracking, and to
266 * assign blame.
267 */
268 struct drm_i915_file_private *file;
Ben Widawsky0260c422014-03-22 22:47:21 -0700269 struct list_head global_link;
Mika Kuoppalac44ef602015-06-25 18:35:05 +0300270 u64 total; /* size addr space maps (ex. 2GB for ggtt) */
Weinan Liff8f7972017-05-31 10:35:52 +0800271 u64 reserved; /* size addr space reserved */
Ben Widawsky0260c422014-03-22 22:47:21 -0700272
Chris Wilson50e046b2016-08-04 07:52:46 +0100273 bool closed;
274
Chris Wilson8bcdd0f72016-08-22 08:44:30 +0100275 struct i915_page_dma scratch_page;
Mika Kuoppala79ab9372015-06-25 18:35:17 +0300276 struct i915_page_table *scratch_pt;
277 struct i915_page_directory *scratch_pd;
Michel Thierry69ab76f2015-07-29 17:23:55 +0100278 struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
Ben Widawsky0260c422014-03-22 22:47:21 -0700279
280 /**
281 * List of objects currently involved in rendering.
282 *
283 * Includes buffers having the contents of their GPU caches
John Harrison97b2a6a2014-11-24 18:49:26 +0000284 * flushed, not necessarily primitives. last_read_req
Ben Widawsky0260c422014-03-22 22:47:21 -0700285 * represents when the rendering involved will be completed.
286 *
287 * A reference is held on the buffer while on this list.
288 */
289 struct list_head active_list;
290
291 /**
292 * LRU list of objects which are not in the ringbuffer and
293 * are ready to unbind, but are still in the GTT.
294 *
John Harrison97b2a6a2014-11-24 18:49:26 +0000295 * last_read_req is NULL while an object is in this list.
Ben Widawsky0260c422014-03-22 22:47:21 -0700296 *
297 * A reference is not held on the buffer while on this list,
298 * as merely being GTT-bound shouldn't prevent its being
299 * freed, and we'll pull it off the list in the free path.
300 */
301 struct list_head inactive_list;
302
Chris Wilson50e046b2016-08-04 07:52:46 +0100303 /**
304 * List of vma that have been unbound.
305 *
306 * A reference is not held on the buffer while on this list.
307 */
308 struct list_head unbound_list;
309
Chris Wilson84486612017-02-15 08:43:40 +0000310 struct pagevec free_pages;
311 bool pt_kmap_wc;
312
Ben Widawsky0260c422014-03-22 22:47:21 -0700313 /* FIXME: Need a more generic return type */
Michel Thierry07749ef2015-03-16 16:00:54 +0000314 gen6_pte_t (*pte_encode)(dma_addr_t addr,
315 enum i915_cache_level level,
Michał Winiarski4fb84d92016-10-13 14:02:40 +0200316 u32 flags); /* Create a valid PTE */
Daniel Vetterf329f5f2015-04-14 17:35:15 +0200317 /* flags for pte_encode */
318#define PTE_READ_ONLY (1<<0)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000319 int (*allocate_va_range)(struct i915_address_space *vm,
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000320 u64 start, u64 length);
Ben Widawsky0260c422014-03-22 22:47:21 -0700321 void (*clear_range)(struct i915_address_space *vm,
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000322 u64 start, u64 length);
Chris Wilsond6473f52016-06-10 14:22:59 +0530323 void (*insert_page)(struct i915_address_space *vm,
324 dma_addr_t addr,
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000325 u64 offset,
Chris Wilsond6473f52016-06-10 14:22:59 +0530326 enum i915_cache_level cache_level,
327 u32 flags);
Ben Widawsky0260c422014-03-22 22:47:21 -0700328 void (*insert_entries)(struct i915_address_space *vm,
Matthew Auld4a234c52017-06-22 10:58:36 +0100329 struct i915_vma *vma,
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000330 enum i915_cache_level cache_level,
331 u32 flags);
Ben Widawsky0260c422014-03-22 22:47:21 -0700332 void (*cleanup)(struct i915_address_space *vm);
Daniel Vetter777dc5b2015-04-14 17:35:12 +0200333 /** Unmap an object from an address space. This usually consists of
334 * setting the valid PTE entries to a reserved scratch page. */
335 void (*unbind_vma)(struct i915_vma *vma);
336 /* Map an object into an address space with the given cache flags. */
Daniel Vetter70b9f6f2015-04-14 17:35:27 +0200337 int (*bind_vma)(struct i915_vma *vma,
338 enum i915_cache_level cache_level,
339 u32 flags);
Matthew Auldfa3f46a2017-10-06 23:18:19 +0100340 int (*set_pages)(struct i915_vma *vma);
341 void (*clear_pages)(struct i915_vma *vma);
Chris Wilson84486612017-02-15 08:43:40 +0000342
343 I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
Ben Widawsky0260c422014-03-22 22:47:21 -0700344};
345
Chris Wilson2bfa9962016-08-04 07:52:25 +0100346#define i915_is_ggtt(V) (!(V)->file)
Chris Wilson596c5922016-02-26 11:03:20 +0000347
Mika Kuoppala3e490042017-02-28 17:28:07 +0200348static inline bool
349i915_vm_is_48bit(const struct i915_address_space *vm)
350{
351 return (vm->total - 1) >> 32;
352}
353
Ben Widawsky0260c422014-03-22 22:47:21 -0700354/* The Graphics Translation Table is the way in which GEN hardware translates a
355 * Graphics Virtual Address into a Physical Address. In addition to the normal
356 * collateral associated with any va->pa translations GEN hardware also has a
357 * portion of the GTT which can be mapped by the CPU and remain both coherent
358 * and correct (in cases like swizzling). That region is referred to as GMADR in
359 * the spec.
360 */
Joonas Lahtinen62106b42016-03-18 10:42:57 +0200361struct i915_ggtt {
Ben Widawsky0260c422014-03-22 22:47:21 -0700362 struct i915_address_space base;
Chris Wilsonf7bbe782016-08-19 16:54:27 +0100363 struct io_mapping mappable; /* Mapping to our CPU mappable region */
Ben Widawsky0260c422014-03-22 22:47:21 -0700364
Chris Wilsonedd1f2f2017-01-06 15:20:11 +0000365 phys_addr_t mappable_base; /* PA of our GMADR */
366 u64 mappable_end; /* End offset that we can CPU map */
367
Paulo Zanoni3c6b29b2016-12-15 11:23:55 -0200368 /* Stolen memory is segmented in hardware with different portions
369 * offlimits to certain functions.
370 *
371 * The drm_mm is initialised to the total accessible range, as found
372 * from the PCI config. On Broadwell+, this is further restricted to
373 * avoid the first page! The upper end of stolen memory is reserved for
374 * hardware functions and similarly removed from the accessible range.
375 */
Chris Wilsonedd1f2f2017-01-06 15:20:11 +0000376 u32 stolen_size; /* Total size of stolen memory */
377 u32 stolen_usable_size; /* Total size minus reserved ranges */
378 u32 stolen_reserved_base;
379 u32 stolen_reserved_size;
Ben Widawsky0260c422014-03-22 22:47:21 -0700380
381 /** "Graphics Stolen Memory" holds the global PTEs */
382 void __iomem *gsm;
Chris Wilson7c3f86b2017-01-12 11:00:49 +0000383 void (*invalidate)(struct drm_i915_private *dev_priv);
Ben Widawsky0260c422014-03-22 22:47:21 -0700384
385 bool do_idle_maps;
386
387 int mtrr;
Chris Wilson95374d72016-10-12 10:05:20 +0100388
389 struct drm_mm_node error_capture;
Ben Widawsky0260c422014-03-22 22:47:21 -0700390};
391
392struct i915_hw_ppgtt {
393 struct i915_address_space base;
394 struct kref ref;
395 struct drm_mm_node node;
Ben Widawsky563222a2015-03-19 12:53:28 +0000396 unsigned long pd_dirty_rings;
Ben Widawsky0260c422014-03-22 22:47:21 -0700397 union {
Michel Thierry81ba8aef2015-08-03 09:52:01 +0100398 struct i915_pml4 pml4; /* GEN8+ & 48b PPGTT */
399 struct i915_page_directory_pointer pdp; /* GEN8+ */
400 struct i915_page_directory pd; /* GEN6-7 */
Ben Widawskyd7b3de92015-02-24 16:22:34 +0000401 };
Ben Widawsky0260c422014-03-22 22:47:21 -0700402
Ben Widawsky678d96f2015-03-16 16:00:56 +0000403 gen6_pte_t __iomem *pd_addr;
404
Ben Widawsky0260c422014-03-22 22:47:21 -0700405 int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
John Harrisone85b26d2015-05-29 17:43:56 +0100406 struct drm_i915_gem_request *req);
Ben Widawsky0260c422014-03-22 22:47:21 -0700407 void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
408};
409
Dave Gordon731f74c2016-06-24 19:37:46 +0100410/*
411 * gen6_for_each_pde() iterates over every pde from start until start+length.
412 * If start and start+length are not perfectly divisible, the macro will round
413 * down and up as needed. Start=0 and length=2G effectively iterates over
414 * every PDE in the system. The macro modifies ALL its parameters except 'pd',
415 * so each of the other parameters should preferably be a simple variable, or
416 * at most an lvalue with no side-effects!
Ben Widawsky678d96f2015-03-16 16:00:56 +0000417 */
Dave Gordon731f74c2016-06-24 19:37:46 +0100418#define gen6_for_each_pde(pt, pd, start, length, iter) \
419 for (iter = gen6_pde_index(start); \
420 length > 0 && iter < I915_PDES && \
421 (pt = (pd)->page_table[iter], true); \
422 ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT); \
423 temp = min(temp - start, length); \
424 start += temp, length -= temp; }), ++iter)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000425
Dave Gordon731f74c2016-06-24 19:37:46 +0100426#define gen6_for_all_pdes(pt, pd, iter) \
427 for (iter = 0; \
428 iter < I915_PDES && \
429 (pt = (pd)->page_table[iter], true); \
430 ++iter)
Michel Thierry09942c62015-04-08 12:13:30 +0100431
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000432static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000433{
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000434 const u32 mask = NUM_PTE(pde_shift) - 1;
Ben Widawsky678d96f2015-03-16 16:00:56 +0000435
436 return (address >> PAGE_SHIFT) & mask;
437}
438
439/* Helper to counts the number of PTEs within the given length. This count
440 * does not cross a page table boundary, so the max value would be
441 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
442*/
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000443static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000444{
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000445 const u64 mask = ~((1ULL << pde_shift) - 1);
446 u64 end;
Ben Widawsky678d96f2015-03-16 16:00:56 +0000447
448 WARN_ON(length == 0);
449 WARN_ON(offset_in_page(addr|length));
450
451 end = addr + length;
452
453 if ((addr & mask) != (end & mask))
454 return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
455
456 return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
457}
458
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000459static inline u32 i915_pde_index(u64 addr, u32 shift)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000460{
461 return (addr >> shift) & I915_PDE_MASK;
462}
463
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000464static inline u32 gen6_pte_index(u32 addr)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000465{
466 return i915_pte_index(addr, GEN6_PDE_SHIFT);
467}
468
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000469static inline u32 gen6_pte_count(u32 addr, u32 length)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000470{
471 return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
472}
473
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000474static inline u32 gen6_pde_index(u32 addr)
Ben Widawsky678d96f2015-03-16 16:00:56 +0000475{
476 return i915_pde_index(addr, GEN6_PDE_SHIFT);
477}
478
Mika Kuoppala3e490042017-02-28 17:28:07 +0200479static inline unsigned int
480i915_pdpes_per_pdp(const struct i915_address_space *vm)
481{
482 if (i915_vm_is_48bit(vm))
483 return GEN8_PML4ES_PER_PML4;
484
Mika Kuoppalae7167762017-02-28 17:28:10 +0200485 return GEN8_3LVL_PDPES;
Mika Kuoppala3e490042017-02-28 17:28:07 +0200486}
487
Michel Thierry9271d952015-04-08 12:13:26 +0100488/* Equivalent to the gen6 version, For each pde iterates over every pde
489 * between from start until start + length. On gen8+ it simply iterates
490 * over every page directory entry in a page directory.
491 */
Dave Gordone8ebd8e2015-12-08 13:30:51 +0000492#define gen8_for_each_pde(pt, pd, start, length, iter) \
493 for (iter = gen8_pde_index(start); \
494 length > 0 && iter < I915_PDES && \
495 (pt = (pd)->page_table[iter], true); \
496 ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT); \
497 temp = min(temp - start, length); \
498 start += temp, length -= temp; }), ++iter)
Michel Thierry9271d952015-04-08 12:13:26 +0100499
Dave Gordone8ebd8e2015-12-08 13:30:51 +0000500#define gen8_for_each_pdpe(pd, pdp, start, length, iter) \
501 for (iter = gen8_pdpe_index(start); \
Mika Kuoppala3e490042017-02-28 17:28:07 +0200502 length > 0 && iter < i915_pdpes_per_pdp(vm) && \
Dave Gordone8ebd8e2015-12-08 13:30:51 +0000503 (pd = (pdp)->page_directory[iter], true); \
504 ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT); \
505 temp = min(temp - start, length); \
506 start += temp, length -= temp; }), ++iter)
Michel Thierry9271d952015-04-08 12:13:26 +0100507
Dave Gordone8ebd8e2015-12-08 13:30:51 +0000508#define gen8_for_each_pml4e(pdp, pml4, start, length, iter) \
509 for (iter = gen8_pml4e_index(start); \
510 length > 0 && iter < GEN8_PML4ES_PER_PML4 && \
511 (pdp = (pml4)->pdps[iter], true); \
512 ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT); \
513 temp = min(temp - start, length); \
514 start += temp, length -= temp; }), ++iter)
Michel Thierry762d9932015-07-30 11:05:29 +0100515
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000516static inline u32 gen8_pte_index(u64 address)
Michel Thierry9271d952015-04-08 12:13:26 +0100517{
518 return i915_pte_index(address, GEN8_PDE_SHIFT);
519}
520
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000521static inline u32 gen8_pde_index(u64 address)
Michel Thierry9271d952015-04-08 12:13:26 +0100522{
523 return i915_pde_index(address, GEN8_PDE_SHIFT);
524}
525
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000526static inline u32 gen8_pdpe_index(u64 address)
Michel Thierry9271d952015-04-08 12:13:26 +0100527{
528 return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
529}
530
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000531static inline u32 gen8_pml4e_index(u64 address)
Michel Thierry9271d952015-04-08 12:13:26 +0100532{
Michel Thierry762d9932015-07-30 11:05:29 +0100533 return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
Michel Thierry9271d952015-04-08 12:13:26 +0100534}
535
Chris Wilson75c7b0b2017-02-15 08:43:57 +0000536static inline u64 gen8_pte_count(u64 address, u64 length)
Michel Thierry33c88192015-04-08 12:13:33 +0100537{
538 return i915_pte_count(address, length, GEN8_PDE_SHIFT);
539}
540
Mika Kuoppalad852c7b2015-06-25 18:35:06 +0300541static inline dma_addr_t
542i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
543{
Chris Wilsonfe52e372017-02-15 08:43:47 +0000544 return px_dma(ppgtt->pdp.page_directory[n]);
Mika Kuoppalad852c7b2015-06-25 18:35:06 +0300545}
546
Joonas Lahtinenb42fe9c2016-11-11 12:43:54 +0200547static inline struct i915_ggtt *
548i915_vm_to_ggtt(struct i915_address_space *vm)
549{
550 GEM_BUG_ON(!i915_is_ggtt(vm));
551 return container_of(vm, struct i915_ggtt, base);
552}
553
Zhi Wang43958902017-09-14 20:39:40 +0800554#define INTEL_MAX_PPAT_ENTRIES 8
555#define INTEL_PPAT_PERFECT_MATCH (~0U)
556
557struct intel_ppat;
558
559struct intel_ppat_entry {
560 struct intel_ppat *ppat;
561 struct kref ref;
562 u8 value;
563};
564
565struct intel_ppat {
566 struct intel_ppat_entry entries[INTEL_MAX_PPAT_ENTRIES];
567 DECLARE_BITMAP(used, INTEL_MAX_PPAT_ENTRIES);
568 DECLARE_BITMAP(dirty, INTEL_MAX_PPAT_ENTRIES);
569 unsigned int max_entries;
570 u8 clear_value;
571 /*
572 * Return a score to show how two PPAT values match,
573 * a INTEL_PPAT_PERFECT_MATCH indicates a perfect match
574 */
575 unsigned int (*match)(u8 src, u8 dst);
576 void (*update_hw)(struct drm_i915_private *i915);
577
578 struct drm_i915_private *i915;
579};
580
581const struct intel_ppat_entry *
582intel_ppat_get(struct drm_i915_private *i915, u8 value);
583void intel_ppat_put(const struct intel_ppat_entry *entry);
584
Chris Wilson6cde9a02017-02-13 17:15:50 +0000585int i915_gem_init_aliasing_ppgtt(struct drm_i915_private *i915);
586void i915_gem_fini_aliasing_ppgtt(struct drm_i915_private *i915);
587
Chris Wilson97d6d7a2016-08-04 07:52:22 +0100588int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv);
589int i915_ggtt_init_hw(struct drm_i915_private *dev_priv);
590int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv);
Chris Wilson7c3f86b2017-01-12 11:00:49 +0000591void i915_ggtt_enable_guc(struct drm_i915_private *i915);
592void i915_ggtt_disable_guc(struct drm_i915_private *i915);
Chris Wilsonf6b9d5c2016-08-04 07:52:23 +0100593int i915_gem_init_ggtt(struct drm_i915_private *dev_priv);
Chris Wilson97d6d7a2016-08-04 07:52:22 +0100594void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv);
Daniel Vetteree960be2014-08-06 15:04:45 +0200595
Tvrtko Ursulinc6be6072016-11-16 08:55:31 +0000596int i915_ppgtt_init_hw(struct drm_i915_private *dev_priv);
Daniel Vetteree960be2014-08-06 15:04:45 +0200597void i915_ppgtt_release(struct kref *kref);
Chris Wilson2bfa9962016-08-04 07:52:25 +0100598struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv,
Chris Wilson80b204b2016-10-28 13:58:58 +0100599 struct drm_i915_file_private *fpriv,
600 const char *name);
Chris Wilson0c7eeda2017-01-11 21:09:25 +0000601void i915_ppgtt_close(struct i915_address_space *vm);
Daniel Vetteree960be2014-08-06 15:04:45 +0200602static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
603{
604 if (ppgtt)
605 kref_get(&ppgtt->ref);
606}
607static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
608{
609 if (ppgtt)
610 kref_put(&ppgtt->ref, i915_ppgtt_release);
611}
Ben Widawsky0260c422014-03-22 22:47:21 -0700612
Chris Wilsondc979972016-05-10 14:10:04 +0100613void i915_check_and_clear_faults(struct drm_i915_private *dev_priv);
Tvrtko Ursulin275a9912016-11-16 08:55:34 +0000614void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv);
615void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv);
Ben Widawsky0260c422014-03-22 22:47:21 -0700616
Chris Wilson03ac84f2016-10-28 13:58:36 +0100617int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
618 struct sg_table *pages);
619void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
620 struct sg_table *pages);
Ben Widawsky0260c422014-03-22 22:47:21 -0700621
Chris Wilson625d9882017-01-11 11:23:11 +0000622int i915_gem_gtt_reserve(struct i915_address_space *vm,
623 struct drm_mm_node *node,
624 u64 size, u64 offset, unsigned long color,
625 unsigned int flags);
626
Chris Wilsone007b192017-01-11 11:23:10 +0000627int i915_gem_gtt_insert(struct i915_address_space *vm,
628 struct drm_mm_node *node,
629 u64 size, u64 alignment, unsigned long color,
630 u64 start, u64 end, unsigned int flags);
631
Chris Wilson59bfa122016-08-04 16:32:31 +0100632/* Flags used by pin/bind&friends. */
Chris Wilson305bc232016-08-04 16:32:33 +0100633#define PIN_NONBLOCK BIT(0)
634#define PIN_MAPPABLE BIT(1)
635#define PIN_ZONE_4G BIT(2)
Chris Wilson82118872016-08-18 17:17:05 +0100636#define PIN_NONFAULT BIT(3)
Chris Wilson616d9ce2017-06-16 15:05:21 +0100637#define PIN_NOEVICT BIT(4)
Chris Wilson305bc232016-08-04 16:32:33 +0100638
639#define PIN_MBZ BIT(5) /* I915_VMA_PIN_OVERFLOW */
640#define PIN_GLOBAL BIT(6) /* I915_VMA_GLOBAL_BIND */
641#define PIN_USER BIT(7) /* I915_VMA_LOCAL_BIND */
642#define PIN_UPDATE BIT(8)
643
644#define PIN_HIGH BIT(9)
645#define PIN_OFFSET_BIAS BIT(10)
646#define PIN_OFFSET_FIXED BIT(11)
Chris Wilsonf51455d2017-01-10 14:47:34 +0000647#define PIN_OFFSET_MASK (-I915_GTT_PAGE_SIZE)
Chris Wilson59bfa122016-08-04 16:32:31 +0100648
Ben Widawsky0260c422014-03-22 22:47:21 -0700649#endif