Rev 5354 | Rev 6088 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 5354 | Rev 6084 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright © 2014 Intel Corporation |
2 | * Copyright © 2014 Intel Corporation |
3 | * |
3 | * |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * copy of this software and associated documentation files (the "Software"), |
5 | * copy of this software and associated documentation files (the "Software"), |
6 | * to deal in the Software without restriction, including without limitation |
6 | * to deal in the Software without restriction, including without limitation |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
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 |
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: |
9 | * Software is furnished to do so, subject to the following conditions: |
10 | * |
10 | * |
11 | * The above copyright notice and this permission notice (including the next |
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 |
12 | * paragraph) shall be included in all copies or substantial portions of the |
13 | * Software. |
13 | * Software. |
14 | * |
14 | * |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
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, |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
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 |
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 |
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 |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
21 | * IN THE SOFTWARE. |
21 | * IN THE SOFTWARE. |
22 | * |
22 | * |
23 | * Authors: |
23 | * Authors: |
24 | * Ben Widawsky |
24 | * Ben Widawsky |
25 | * Michel Thierry |
25 | * Michel Thierry |
26 | * Thomas Daniel |
26 | * Thomas Daniel |
27 | * Oscar Mateo |
27 | * Oscar Mateo |
28 | * |
28 | * |
29 | */ |
29 | */ |
30 | 30 | ||
31 | /** |
31 | /** |
32 | * DOC: Logical Rings, Logical Ring Contexts and Execlists |
32 | * DOC: Logical Rings, Logical Ring Contexts and Execlists |
33 | * |
33 | * |
34 | * Motivation: |
34 | * Motivation: |
35 | * GEN8 brings an expansion of the HW contexts: "Logical Ring Contexts". |
35 | * GEN8 brings an expansion of the HW contexts: "Logical Ring Contexts". |
36 | * These expanded contexts enable a number of new abilities, especially |
36 | * These expanded contexts enable a number of new abilities, especially |
37 | * "Execlists" (also implemented in this file). |
37 | * "Execlists" (also implemented in this file). |
38 | * |
38 | * |
39 | * One of the main differences with the legacy HW contexts is that logical |
39 | * One of the main differences with the legacy HW contexts is that logical |
40 | * ring contexts incorporate many more things to the context's state, like |
40 | * ring contexts incorporate many more things to the context's state, like |
41 | * PDPs or ringbuffer control registers: |
41 | * PDPs or ringbuffer control registers: |
42 | * |
42 | * |
43 | * The reason why PDPs are included in the context is straightforward: as |
43 | * The reason why PDPs are included in the context is straightforward: as |
44 | * PPGTTs (per-process GTTs) are actually per-context, having the PDPs |
44 | * PPGTTs (per-process GTTs) are actually per-context, having the PDPs |
45 | * contained there mean you don't need to do a ppgtt->switch_mm yourself, |
45 | * contained there mean you don't need to do a ppgtt->switch_mm yourself, |
46 | * instead, the GPU will do it for you on the context switch. |
46 | * instead, the GPU will do it for you on the context switch. |
47 | * |
47 | * |
48 | * But, what about the ringbuffer control registers (head, tail, etc..)? |
48 | * But, what about the ringbuffer control registers (head, tail, etc..)? |
49 | * shouldn't we just need a set of those per engine command streamer? This is |
49 | * shouldn't we just need a set of those per engine command streamer? This is |
50 | * where the name "Logical Rings" starts to make sense: by virtualizing the |
50 | * where the name "Logical Rings" starts to make sense: by virtualizing the |
51 | * rings, the engine cs shifts to a new "ring buffer" with every context |
51 | * rings, the engine cs shifts to a new "ring buffer" with every context |
52 | * switch. When you want to submit a workload to the GPU you: A) choose your |
52 | * switch. When you want to submit a workload to the GPU you: A) choose your |
53 | * context, B) find its appropriate virtualized ring, C) write commands to it |
53 | * context, B) find its appropriate virtualized ring, C) write commands to it |
54 | * and then, finally, D) tell the GPU to switch to that context. |
54 | * and then, finally, D) tell the GPU to switch to that context. |
55 | * |
55 | * |
56 | * Instead of the legacy MI_SET_CONTEXT, the way you tell the GPU to switch |
56 | * Instead of the legacy MI_SET_CONTEXT, the way you tell the GPU to switch |
57 | * to a contexts is via a context execution list, ergo "Execlists". |
57 | * to a contexts is via a context execution list, ergo "Execlists". |
58 | * |
58 | * |
59 | * LRC implementation: |
59 | * LRC implementation: |
60 | * Regarding the creation of contexts, we have: |
60 | * Regarding the creation of contexts, we have: |
61 | * |
61 | * |
62 | * - One global default context. |
62 | * - One global default context. |
63 | * - One local default context for each opened fd. |
63 | * - One local default context for each opened fd. |
64 | * - One local extra context for each context create ioctl call. |
64 | * - One local extra context for each context create ioctl call. |
65 | * |
65 | * |
66 | * Now that ringbuffers belong per-context (and not per-engine, like before) |
66 | * Now that ringbuffers belong per-context (and not per-engine, like before) |
67 | * and that contexts are uniquely tied to a given engine (and not reusable, |
67 | * and that contexts are uniquely tied to a given engine (and not reusable, |
68 | * like before) we need: |
68 | * like before) we need: |
69 | * |
69 | * |
70 | * - One ringbuffer per-engine inside each context. |
70 | * - One ringbuffer per-engine inside each context. |
71 | * - One backing object per-engine inside each context. |
71 | * - One backing object per-engine inside each context. |
72 | * |
72 | * |
73 | * The global default context starts its life with these new objects fully |
73 | * The global default context starts its life with these new objects fully |
74 | * allocated and populated. The local default context for each opened fd is |
74 | * allocated and populated. The local default context for each opened fd is |
75 | * more complex, because we don't know at creation time which engine is going |
75 | * more complex, because we don't know at creation time which engine is going |
76 | * to use them. To handle this, we have implemented a deferred creation of LR |
76 | * to use them. To handle this, we have implemented a deferred creation of LR |
77 | * contexts: |
77 | * contexts: |
78 | * |
78 | * |
79 | * The local context starts its life as a hollow or blank holder, that only |
79 | * The local context starts its life as a hollow or blank holder, that only |
80 | * gets populated for a given engine once we receive an execbuffer. If later |
80 | * gets populated for a given engine once we receive an execbuffer. If later |
81 | * on we receive another execbuffer ioctl for the same context but a different |
81 | * on we receive another execbuffer ioctl for the same context but a different |
82 | * engine, we allocate/populate a new ringbuffer and context backing object and |
82 | * engine, we allocate/populate a new ringbuffer and context backing object and |
83 | * so on. |
83 | * so on. |
84 | * |
84 | * |
85 | * Finally, regarding local contexts created using the ioctl call: as they are |
85 | * Finally, regarding local contexts created using the ioctl call: as they are |
86 | * only allowed with the render ring, we can allocate & populate them right |
86 | * only allowed with the render ring, we can allocate & populate them right |
87 | * away (no need to defer anything, at least for now). |
87 | * away (no need to defer anything, at least for now). |
88 | * |
88 | * |
89 | * Execlists implementation: |
89 | * Execlists implementation: |
90 | * Execlists are the new method by which, on gen8+ hardware, workloads are |
90 | * Execlists are the new method by which, on gen8+ hardware, workloads are |
91 | * submitted for execution (as opposed to the legacy, ringbuffer-based, method). |
91 | * submitted for execution (as opposed to the legacy, ringbuffer-based, method). |
92 | * This method works as follows: |
92 | * This method works as follows: |
93 | * |
93 | * |
94 | * When a request is committed, its commands (the BB start and any leading or |
94 | * When a request is committed, its commands (the BB start and any leading or |
95 | * trailing commands, like the seqno breadcrumbs) are placed in the ringbuffer |
95 | * trailing commands, like the seqno breadcrumbs) are placed in the ringbuffer |
96 | * for the appropriate context. The tail pointer in the hardware context is not |
96 | * for the appropriate context. The tail pointer in the hardware context is not |
97 | * updated at this time, but instead, kept by the driver in the ringbuffer |
97 | * updated at this time, but instead, kept by the driver in the ringbuffer |
98 | * structure. A structure representing this request is added to a request queue |
98 | * structure. A structure representing this request is added to a request queue |
99 | * for the appropriate engine: this structure contains a copy of the context's |
99 | * for the appropriate engine: this structure contains a copy of the context's |
100 | * tail after the request was written to the ring buffer and a pointer to the |
100 | * tail after the request was written to the ring buffer and a pointer to the |
101 | * context itself. |
101 | * context itself. |
102 | * |
102 | * |
103 | * If the engine's request queue was empty before the request was added, the |
103 | * If the engine's request queue was empty before the request was added, the |
104 | * queue is processed immediately. Otherwise the queue will be processed during |
104 | * queue is processed immediately. Otherwise the queue will be processed during |
105 | * a context switch interrupt. In any case, elements on the queue will get sent |
105 | * a context switch interrupt. In any case, elements on the queue will get sent |
106 | * (in pairs) to the GPU's ExecLists Submit Port (ELSP, for short) with a |
106 | * (in pairs) to the GPU's ExecLists Submit Port (ELSP, for short) with a |
107 | * globally unique 20-bits submission ID. |
107 | * globally unique 20-bits submission ID. |
108 | * |
108 | * |
109 | * When execution of a request completes, the GPU updates the context status |
109 | * When execution of a request completes, the GPU updates the context status |
110 | * buffer with a context complete event and generates a context switch interrupt. |
110 | * buffer with a context complete event and generates a context switch interrupt. |
111 | * During the interrupt handling, the driver examines the events in the buffer: |
111 | * During the interrupt handling, the driver examines the events in the buffer: |
112 | * for each context complete event, if the announced ID matches that on the head |
112 | * for each context complete event, if the announced ID matches that on the head |
113 | * of the request queue, then that request is retired and removed from the queue. |
113 | * of the request queue, then that request is retired and removed from the queue. |
114 | * |
114 | * |
115 | * After processing, if any requests were retired and the queue is not empty |
115 | * After processing, if any requests were retired and the queue is not empty |
116 | * then a new execution list can be submitted. The two requests at the front of |
116 | * then a new execution list can be submitted. The two requests at the front of |
117 | * the queue are next to be submitted but since a context may not occur twice in |
117 | * the queue are next to be submitted but since a context may not occur twice in |
118 | * an execution list, if subsequent requests have the same ID as the first then |
118 | * an execution list, if subsequent requests have the same ID as the first then |
119 | * the two requests must be combined. This is done simply by discarding requests |
119 | * the two requests must be combined. This is done simply by discarding requests |
120 | * at the head of the queue until either only one requests is left (in which case |
120 | * at the head of the queue until either only one requests is left (in which case |
121 | * we use a NULL second context) or the first two requests have unique IDs. |
121 | * we use a NULL second context) or the first two requests have unique IDs. |
122 | * |
122 | * |
123 | * By always executing the first two requests in the queue the driver ensures |
123 | * By always executing the first two requests in the queue the driver ensures |
124 | * that the GPU is kept as busy as possible. In the case where a single context |
124 | * that the GPU is kept as busy as possible. In the case where a single context |
125 | * completes but a second context is still executing, the request for this second |
125 | * completes but a second context is still executing, the request for this second |
126 | * context will be at the head of the queue when we remove the first one. This |
126 | * context will be at the head of the queue when we remove the first one. This |
127 | * request will then be resubmitted along with a new request for a different context, |
127 | * request will then be resubmitted along with a new request for a different context, |
128 | * which will cause the hardware to continue executing the second request and queue |
128 | * which will cause the hardware to continue executing the second request and queue |
129 | * the new request (the GPU detects the condition of a context getting preempted |
129 | * the new request (the GPU detects the condition of a context getting preempted |
130 | * with the same context and optimizes the context switch flow by not doing |
130 | * with the same context and optimizes the context switch flow by not doing |
131 | * preemption, but just sampling the new tail pointer). |
131 | * preemption, but just sampling the new tail pointer). |
132 | * |
132 | * |
133 | */ |
133 | */ |
134 | 134 | ||
135 | #include |
135 | #include |
136 | #include |
136 | #include |
137 | #include "intel_drv.h" |
137 | #include "intel_drv.h" |
138 | #include "i915_drv.h" |
138 | #include "intel_mocs.h" |
139 | 139 | ||
140 | #define GEN9_LR_CONTEXT_RENDER_SIZE (22 * PAGE_SIZE) |
140 | #define GEN9_LR_CONTEXT_RENDER_SIZE (22 * PAGE_SIZE) |
141 | #define GEN8_LR_CONTEXT_RENDER_SIZE (20 * PAGE_SIZE) |
141 | #define GEN8_LR_CONTEXT_RENDER_SIZE (20 * PAGE_SIZE) |
142 | #define GEN8_LR_CONTEXT_OTHER_SIZE (2 * PAGE_SIZE) |
142 | #define GEN8_LR_CONTEXT_OTHER_SIZE (2 * PAGE_SIZE) |
143 | 143 | ||
144 | #define RING_EXECLIST_QFULL (1 << 0x2) |
144 | #define RING_EXECLIST_QFULL (1 << 0x2) |
145 | #define RING_EXECLIST1_VALID (1 << 0x3) |
145 | #define RING_EXECLIST1_VALID (1 << 0x3) |
146 | #define RING_EXECLIST0_VALID (1 << 0x4) |
146 | #define RING_EXECLIST0_VALID (1 << 0x4) |
147 | #define RING_EXECLIST_ACTIVE_STATUS (3 << 0xE) |
147 | #define RING_EXECLIST_ACTIVE_STATUS (3 << 0xE) |
148 | #define RING_EXECLIST1_ACTIVE (1 << 0x11) |
148 | #define RING_EXECLIST1_ACTIVE (1 << 0x11) |
149 | #define RING_EXECLIST0_ACTIVE (1 << 0x12) |
149 | #define RING_EXECLIST0_ACTIVE (1 << 0x12) |
150 | 150 | ||
151 | #define GEN8_CTX_STATUS_IDLE_ACTIVE (1 << 0) |
151 | #define GEN8_CTX_STATUS_IDLE_ACTIVE (1 << 0) |
152 | #define GEN8_CTX_STATUS_PREEMPTED (1 << 1) |
152 | #define GEN8_CTX_STATUS_PREEMPTED (1 << 1) |
153 | #define GEN8_CTX_STATUS_ELEMENT_SWITCH (1 << 2) |
153 | #define GEN8_CTX_STATUS_ELEMENT_SWITCH (1 << 2) |
154 | #define GEN8_CTX_STATUS_ACTIVE_IDLE (1 << 3) |
154 | #define GEN8_CTX_STATUS_ACTIVE_IDLE (1 << 3) |
155 | #define GEN8_CTX_STATUS_COMPLETE (1 << 4) |
155 | #define GEN8_CTX_STATUS_COMPLETE (1 << 4) |
156 | #define GEN8_CTX_STATUS_LITE_RESTORE (1 << 15) |
156 | #define GEN8_CTX_STATUS_LITE_RESTORE (1 << 15) |
157 | 157 | ||
158 | #define CTX_LRI_HEADER_0 0x01 |
158 | #define CTX_LRI_HEADER_0 0x01 |
159 | #define CTX_CONTEXT_CONTROL 0x02 |
159 | #define CTX_CONTEXT_CONTROL 0x02 |
160 | #define CTX_RING_HEAD 0x04 |
160 | #define CTX_RING_HEAD 0x04 |
161 | #define CTX_RING_TAIL 0x06 |
161 | #define CTX_RING_TAIL 0x06 |
162 | #define CTX_RING_BUFFER_START 0x08 |
162 | #define CTX_RING_BUFFER_START 0x08 |
163 | #define CTX_RING_BUFFER_CONTROL 0x0a |
163 | #define CTX_RING_BUFFER_CONTROL 0x0a |
164 | #define CTX_BB_HEAD_U 0x0c |
164 | #define CTX_BB_HEAD_U 0x0c |
165 | #define CTX_BB_HEAD_L 0x0e |
165 | #define CTX_BB_HEAD_L 0x0e |
166 | #define CTX_BB_STATE 0x10 |
166 | #define CTX_BB_STATE 0x10 |
167 | #define CTX_SECOND_BB_HEAD_U 0x12 |
167 | #define CTX_SECOND_BB_HEAD_U 0x12 |
168 | #define CTX_SECOND_BB_HEAD_L 0x14 |
168 | #define CTX_SECOND_BB_HEAD_L 0x14 |
169 | #define CTX_SECOND_BB_STATE 0x16 |
169 | #define CTX_SECOND_BB_STATE 0x16 |
170 | #define CTX_BB_PER_CTX_PTR 0x18 |
170 | #define CTX_BB_PER_CTX_PTR 0x18 |
171 | #define CTX_RCS_INDIRECT_CTX 0x1a |
171 | #define CTX_RCS_INDIRECT_CTX 0x1a |
172 | #define CTX_RCS_INDIRECT_CTX_OFFSET 0x1c |
172 | #define CTX_RCS_INDIRECT_CTX_OFFSET 0x1c |
173 | #define CTX_LRI_HEADER_1 0x21 |
173 | #define CTX_LRI_HEADER_1 0x21 |
174 | #define CTX_CTX_TIMESTAMP 0x22 |
174 | #define CTX_CTX_TIMESTAMP 0x22 |
175 | #define CTX_PDP3_UDW 0x24 |
175 | #define CTX_PDP3_UDW 0x24 |
176 | #define CTX_PDP3_LDW 0x26 |
176 | #define CTX_PDP3_LDW 0x26 |
177 | #define CTX_PDP2_UDW 0x28 |
177 | #define CTX_PDP2_UDW 0x28 |
178 | #define CTX_PDP2_LDW 0x2a |
178 | #define CTX_PDP2_LDW 0x2a |
179 | #define CTX_PDP1_UDW 0x2c |
179 | #define CTX_PDP1_UDW 0x2c |
180 | #define CTX_PDP1_LDW 0x2e |
180 | #define CTX_PDP1_LDW 0x2e |
181 | #define CTX_PDP0_UDW 0x30 |
181 | #define CTX_PDP0_UDW 0x30 |
182 | #define CTX_PDP0_LDW 0x32 |
182 | #define CTX_PDP0_LDW 0x32 |
183 | #define CTX_LRI_HEADER_2 0x41 |
183 | #define CTX_LRI_HEADER_2 0x41 |
184 | #define CTX_R_PWR_CLK_STATE 0x42 |
184 | #define CTX_R_PWR_CLK_STATE 0x42 |
185 | #define CTX_GPGPU_CSR_BASE_ADDRESS 0x44 |
185 | #define CTX_GPGPU_CSR_BASE_ADDRESS 0x44 |
186 | 186 | ||
187 | #define GEN8_CTX_VALID (1<<0) |
187 | #define GEN8_CTX_VALID (1<<0) |
188 | #define GEN8_CTX_FORCE_PD_RESTORE (1<<1) |
188 | #define GEN8_CTX_FORCE_PD_RESTORE (1<<1) |
189 | #define GEN8_CTX_FORCE_RESTORE (1<<2) |
189 | #define GEN8_CTX_FORCE_RESTORE (1<<2) |
190 | #define GEN8_CTX_L3LLC_COHERENT (1<<5) |
190 | #define GEN8_CTX_L3LLC_COHERENT (1<<5) |
191 | #define GEN8_CTX_PRIVILEGE (1<<8) |
191 | #define GEN8_CTX_PRIVILEGE (1<<8) |
- | 192 | ||
- | 193 | #define ASSIGN_CTX_PDP(ppgtt, reg_state, n) { \ |
|
- | 194 | const u64 _addr = i915_page_dir_dma_addr((ppgtt), (n)); \ |
|
- | 195 | reg_state[CTX_PDP ## n ## _UDW+1] = upper_32_bits(_addr); \ |
|
- | 196 | reg_state[CTX_PDP ## n ## _LDW+1] = lower_32_bits(_addr); \ |
|
- | 197 | } |
|
- | 198 | ||
- | 199 | #define ASSIGN_CTX_PML4(ppgtt, reg_state) { \ |
|
- | 200 | reg_state[CTX_PDP0_UDW + 1] = upper_32_bits(px_dma(&ppgtt->pml4)); \ |
|
- | 201 | reg_state[CTX_PDP0_LDW + 1] = lower_32_bits(px_dma(&ppgtt->pml4)); \ |
|
- | 202 | } |
|
- | 203 | ||
192 | enum { |
204 | enum { |
193 | ADVANCED_CONTEXT = 0, |
205 | ADVANCED_CONTEXT = 0, |
194 | LEGACY_CONTEXT, |
206 | LEGACY_32B_CONTEXT, |
195 | ADVANCED_AD_CONTEXT, |
207 | ADVANCED_AD_CONTEXT, |
196 | LEGACY_64B_CONTEXT |
208 | LEGACY_64B_CONTEXT |
197 | }; |
209 | }; |
198 | #define GEN8_CTX_MODE_SHIFT 3 |
210 | #define GEN8_CTX_ADDRESSING_MODE_SHIFT 3 |
- | 211 | #define GEN8_CTX_ADDRESSING_MODE(dev) (USES_FULL_48BIT_PPGTT(dev) ?\ |
|
- | 212 | LEGACY_64B_CONTEXT :\ |
|
- | 213 | LEGACY_32B_CONTEXT) |
|
199 | enum { |
214 | enum { |
200 | FAULT_AND_HANG = 0, |
215 | FAULT_AND_HANG = 0, |
201 | FAULT_AND_HALT, /* Debug only */ |
216 | FAULT_AND_HALT, /* Debug only */ |
202 | FAULT_AND_STREAM, |
217 | FAULT_AND_STREAM, |
203 | FAULT_AND_CONTINUE /* Unsupported */ |
218 | FAULT_AND_CONTINUE /* Unsupported */ |
204 | }; |
219 | }; |
205 | #define GEN8_CTX_ID_SHIFT 32 |
220 | #define GEN8_CTX_ID_SHIFT 32 |
- | 221 | #define CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT 0x17 |
|
- | 222 | ||
- | 223 | static int intel_lr_context_pin(struct drm_i915_gem_request *rq); |
|
- | 224 | static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, |
|
- | 225 | struct drm_i915_gem_object *default_ctx_obj); |
|
206 | - | ||
207 | static int intel_lr_context_pin(struct intel_engine_cs *ring, |
- | |
208 | struct intel_context *ctx); |
226 | |
209 | 227 | ||
210 | /** |
228 | /** |
211 | * intel_sanitize_enable_execlists() - sanitize i915.enable_execlists |
229 | * intel_sanitize_enable_execlists() - sanitize i915.enable_execlists |
212 | * @dev: DRM device. |
230 | * @dev: DRM device. |
213 | * @enable_execlists: value of i915.enable_execlists module parameter. |
231 | * @enable_execlists: value of i915.enable_execlists module parameter. |
214 | * |
232 | * |
215 | * Only certain platforms support Execlists (the prerequisites being |
233 | * Only certain platforms support Execlists (the prerequisites being |
216 | * support for Logical Ring Contexts and Aliasing PPGTT or better), |
234 | * support for Logical Ring Contexts and Aliasing PPGTT or better). |
217 | * and only when enabled via module parameter. |
- | |
218 | * |
235 | * |
219 | * Return: 1 if Execlists is supported and has to be enabled. |
236 | * Return: 1 if Execlists is supported and has to be enabled. |
220 | */ |
237 | */ |
221 | int intel_sanitize_enable_execlists(struct drm_device *dev, int enable_execlists) |
238 | int intel_sanitize_enable_execlists(struct drm_device *dev, int enable_execlists) |
222 | { |
239 | { |
223 | WARN_ON(i915.enable_ppgtt == -1); |
240 | WARN_ON(i915.enable_ppgtt == -1); |
- | 241 | ||
- | 242 | /* On platforms with execlist available, vGPU will only |
|
- | 243 | * support execlist mode, no ring buffer mode. |
|
- | 244 | */ |
|
- | 245 | if (HAS_LOGICAL_RING_CONTEXTS(dev) && intel_vgpu_active(dev)) |
|
- | 246 | return 1; |
|
224 | 247 | ||
225 | if (INTEL_INFO(dev)->gen >= 9) |
248 | if (INTEL_INFO(dev)->gen >= 9) |
226 | return 1; |
249 | return 1; |
227 | 250 | ||
228 | if (enable_execlists == 0) |
251 | if (enable_execlists == 0) |
229 | return 0; |
252 | return 0; |
230 | 253 | ||
231 | if (HAS_LOGICAL_RING_CONTEXTS(dev) && USES_PPGTT(dev) && |
254 | if (HAS_LOGICAL_RING_CONTEXTS(dev) && USES_PPGTT(dev) && |
232 | i915.use_mmio_flip >= 0) |
255 | i915.use_mmio_flip >= 0) |
233 | return 1; |
256 | return 1; |
234 | 257 | ||
235 | return 0; |
258 | return 0; |
236 | } |
259 | } |
237 | 260 | ||
238 | /** |
261 | /** |
239 | * intel_execlists_ctx_id() - get the Execlists Context ID |
262 | * intel_execlists_ctx_id() - get the Execlists Context ID |
240 | * @ctx_obj: Logical Ring Context backing object. |
263 | * @ctx_obj: Logical Ring Context backing object. |
241 | * |
264 | * |
242 | * Do not confuse with ctx->id! Unfortunately we have a name overload |
265 | * Do not confuse with ctx->id! Unfortunately we have a name overload |
243 | * here: the old context ID we pass to userspace as a handler so that |
266 | * here: the old context ID we pass to userspace as a handler so that |
244 | * they can refer to a context, and the new context ID we pass to the |
267 | * they can refer to a context, and the new context ID we pass to the |
245 | * ELSP so that the GPU can inform us of the context status via |
268 | * ELSP so that the GPU can inform us of the context status via |
246 | * interrupts. |
269 | * interrupts. |
247 | * |
270 | * |
248 | * Return: 20-bits globally unique context ID. |
271 | * Return: 20-bits globally unique context ID. |
249 | */ |
272 | */ |
250 | u32 intel_execlists_ctx_id(struct drm_i915_gem_object *ctx_obj) |
273 | u32 intel_execlists_ctx_id(struct drm_i915_gem_object *ctx_obj) |
251 | { |
274 | { |
252 | u32 lrca = i915_gem_obj_ggtt_offset(ctx_obj); |
275 | u32 lrca = i915_gem_obj_ggtt_offset(ctx_obj) + |
- | 276 | LRC_PPHWSP_PN * PAGE_SIZE; |
|
253 | 277 | ||
254 | /* LRCA is required to be 4K aligned so the more significant 20 bits |
278 | /* LRCA is required to be 4K aligned so the more significant 20 bits |
255 | * are globally unique */ |
279 | * are globally unique */ |
256 | return lrca >> 12; |
280 | return lrca >> 12; |
257 | } |
281 | } |
258 | 282 | ||
259 | static uint64_t execlists_ctx_descriptor(struct drm_i915_gem_object *ctx_obj) |
283 | static bool disable_lite_restore_wa(struct intel_engine_cs *ring) |
- | 284 | { |
|
- | 285 | struct drm_device *dev = ring->dev; |
|
- | 286 | ||
- | 287 | return ((IS_SKYLAKE(dev) && INTEL_REVID(dev) <= SKL_REVID_B0) || |
|
- | 288 | (IS_BROXTON(dev) && INTEL_REVID(dev) == BXT_REVID_A0)) && |
|
- | 289 | (ring->id == VCS || ring->id == VCS2); |
|
- | 290 | } |
|
- | 291 | ||
- | 292 | uint64_t intel_lr_context_descriptor(struct intel_context *ctx, |
|
- | 293 | struct intel_engine_cs *ring) |
|
- | 294 | { |
|
260 | { |
295 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; |
261 | uint64_t desc; |
296 | uint64_t desc; |
- | 297 | uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj) + |
|
262 | uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj); |
298 | LRC_PPHWSP_PN * PAGE_SIZE; |
263 | 299 | ||
264 | WARN_ON(lrca & 0xFFFFFFFF00000FFFULL); |
300 | WARN_ON(lrca & 0xFFFFFFFF00000FFFULL); |
265 | 301 | ||
266 | desc = GEN8_CTX_VALID; |
302 | desc = GEN8_CTX_VALID; |
267 | desc |= LEGACY_CONTEXT << GEN8_CTX_MODE_SHIFT; |
303 | desc |= GEN8_CTX_ADDRESSING_MODE(dev) << GEN8_CTX_ADDRESSING_MODE_SHIFT; |
- | 304 | if (IS_GEN8(ctx_obj->base.dev)) |
|
268 | desc |= GEN8_CTX_L3LLC_COHERENT; |
305 | desc |= GEN8_CTX_L3LLC_COHERENT; |
269 | desc |= GEN8_CTX_PRIVILEGE; |
306 | desc |= GEN8_CTX_PRIVILEGE; |
270 | desc |= lrca; |
307 | desc |= lrca; |
271 | desc |= (u64)intel_execlists_ctx_id(ctx_obj) << GEN8_CTX_ID_SHIFT; |
308 | desc |= (u64)intel_execlists_ctx_id(ctx_obj) << GEN8_CTX_ID_SHIFT; |
272 | 309 | ||
273 | /* TODO: WaDisableLiteRestore when we start using semaphore |
310 | /* TODO: WaDisableLiteRestore when we start using semaphore |
274 | * signalling between Command Streamers */ |
311 | * signalling between Command Streamers */ |
275 | /* desc |= GEN8_CTX_FORCE_RESTORE; */ |
312 | /* desc |= GEN8_CTX_FORCE_RESTORE; */ |
- | 313 | ||
- | 314 | /* WaEnableForceRestoreInCtxtDescForVCS:skl */ |
|
- | 315 | /* WaEnableForceRestoreInCtxtDescForVCS:bxt */ |
|
- | 316 | if (disable_lite_restore_wa(ring)) |
|
- | 317 | desc |= GEN8_CTX_FORCE_RESTORE; |
|
276 | 318 | ||
277 | return desc; |
319 | return desc; |
278 | } |
320 | } |
279 | 321 | ||
280 | static void execlists_elsp_write(struct intel_engine_cs *ring, |
- | |
281 | struct drm_i915_gem_object *ctx_obj0, |
322 | static void execlists_elsp_write(struct drm_i915_gem_request *rq0, |
282 | struct drm_i915_gem_object *ctx_obj1) |
323 | struct drm_i915_gem_request *rq1) |
- | 324 | { |
|
- | 325 | ||
283 | { |
326 | struct intel_engine_cs *ring = rq0->ring; |
284 | struct drm_device *dev = ring->dev; |
327 | struct drm_device *dev = ring->dev; |
285 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
286 | uint64_t temp = 0; |
328 | struct drm_i915_private *dev_priv = dev->dev_private; |
287 | uint32_t desc[4]; |
- | |
288 | unsigned long flags; |
- | |
289 | 329 | uint64_t desc[2]; |
|
290 | /* XXX: You must always write both descriptors in the order below. */ |
330 | |
291 | if (ctx_obj1) |
- | |
292 | temp = execlists_ctx_descriptor(ctx_obj1); |
- | |
293 | else |
- | |
294 | temp = 0; |
- | |
295 | desc[1] = (u32)(temp >> 32); |
- | |
296 | desc[0] = (u32)temp; |
- | |
297 | - | ||
298 | temp = execlists_ctx_descriptor(ctx_obj0); |
331 | if (rq1) { |
299 | desc[3] = (u32)(temp >> 32); |
- | |
300 | desc[2] = (u32)temp; |
- | |
301 | - | ||
302 | /* Set Force Wakeup bit to prevent GT from entering C6 while ELSP writes |
- | |
303 | * are in progress. |
- | |
304 | * |
- | |
305 | * The other problem is that we can't just call gen6_gt_force_wake_get() |
- | |
306 | * because that function calls intel_runtime_pm_get(), which might sleep. |
- | |
307 | * Instead, we do the runtime_pm_get/put when creating/destroying requests. |
- | |
308 | */ |
- | |
309 | spin_lock_irqsave(&dev_priv->uncore.lock, flags); |
- | |
310 | if (IS_CHERRYVIEW(dev) || INTEL_INFO(dev)->gen >= 9) { |
- | |
311 | if (dev_priv->uncore.fw_rendercount++ == 0) |
- | |
312 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
- | |
313 | FORCEWAKE_RENDER); |
- | |
314 | if (dev_priv->uncore.fw_mediacount++ == 0) |
- | |
315 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
- | |
316 | FORCEWAKE_MEDIA); |
- | |
317 | if (INTEL_INFO(dev)->gen >= 9) { |
- | |
318 | if (dev_priv->uncore.fw_blittercount++ == 0) |
- | |
319 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
- | |
320 | FORCEWAKE_BLITTER); |
332 | desc[1] = intel_lr_context_descriptor(rq1->ctx, rq1->ring); |
321 | } |
- | |
322 | } else { |
- | |
323 | if (dev_priv->uncore.forcewake_count++ == 0) |
333 | rq1->elsp_submitted++; |
324 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
334 | } else { |
325 | FORCEWAKE_ALL); |
- | |
326 | } |
- | |
327 | spin_unlock_irqrestore(&dev_priv->uncore.lock, flags); |
- | |
328 | - | ||
329 | I915_WRITE(RING_ELSP(ring), desc[1]); |
- | |
330 | I915_WRITE(RING_ELSP(ring), desc[0]); |
- | |
331 | I915_WRITE(RING_ELSP(ring), desc[3]); |
- | |
332 | /* The context is automatically loaded after the following */ |
335 | desc[1] = 0; |
333 | I915_WRITE(RING_ELSP(ring), desc[2]); |
336 | } |
334 | 337 | ||
335 | /* ELSP is a wo register, so use another nearby reg for posting instead */ |
338 | desc[0] = intel_lr_context_descriptor(rq0->ctx, rq0->ring); |
336 | POSTING_READ(RING_EXECLIST_STATUS(ring)); |
- | |
337 | - | ||
338 | /* Release Force Wakeup (see the big comment above). */ |
- | |
339 | spin_lock_irqsave(&dev_priv->uncore.lock, flags); |
- | |
340 | if (IS_CHERRYVIEW(dev) || INTEL_INFO(dev)->gen >= 9) { |
- | |
341 | if (--dev_priv->uncore.fw_rendercount == 0) |
339 | rq0->elsp_submitted++; |
342 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
- | |
343 | FORCEWAKE_RENDER); |
- | |
344 | if (--dev_priv->uncore.fw_mediacount == 0) |
- | |
345 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
340 | |
346 | FORCEWAKE_MEDIA); |
- | |
347 | if (INTEL_INFO(dev)->gen >= 9) { |
- | |
348 | if (--dev_priv->uncore.fw_blittercount == 0) |
- | |
349 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
- | |
350 | FORCEWAKE_BLITTER); |
341 | /* You must always write both descriptors in the order below. */ |
351 | } |
- | |
352 | } else { |
- | |
- | 342 | spin_lock(&dev_priv->uncore.lock); |
|
- | 343 | intel_uncore_forcewake_get__locked(dev_priv, FORCEWAKE_ALL); |
|
- | 344 | I915_WRITE_FW(RING_ELSP(ring), upper_32_bits(desc[1])); |
|
- | 345 | I915_WRITE_FW(RING_ELSP(ring), lower_32_bits(desc[1])); |
|
- | 346 | ||
- | 347 | I915_WRITE_FW(RING_ELSP(ring), upper_32_bits(desc[0])); |
|
- | 348 | /* The context is automatically loaded after the following */ |
|
353 | if (--dev_priv->uncore.forcewake_count == 0) |
349 | I915_WRITE_FW(RING_ELSP(ring), lower_32_bits(desc[0])); |
354 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
350 | |
355 | FORCEWAKE_ALL); |
351 | /* ELSP is a wo register, use another nearby reg for posting */ |
356 | } |
- | |
357 | - | ||
358 | spin_unlock_irqrestore(&dev_priv->uncore.lock, flags); |
352 | POSTING_READ_FW(RING_EXECLIST_STATUS_LO(ring)); |
- | 353 | intel_uncore_forcewake_put__locked(dev_priv, FORCEWAKE_ALL); |
|
- | 354 | spin_unlock(&dev_priv->uncore.lock); |
|
- | 355 | } |
|
- | 356 | ||
359 | } |
357 | static int execlists_update_context(struct drm_i915_gem_request *rq) |
360 | 358 | { |
|
361 | static int execlists_update_context(struct drm_i915_gem_object *ctx_obj, |
359 | struct intel_engine_cs *ring = rq->ring; |
362 | struct drm_i915_gem_object *ring_obj, |
360 | struct i915_hw_ppgtt *ppgtt = rq->ctx->ppgtt; |
363 | u32 tail) |
361 | struct drm_i915_gem_object *ctx_obj = rq->ctx->engine[ring->id].state; |
364 | { |
362 | struct drm_i915_gem_object *rb_obj = rq->ringbuf->obj; |
365 | struct page *page; |
363 | struct page *page; |
366 | uint32_t *reg_state; |
364 | uint32_t *reg_state; |
- | 365 | ||
- | 366 | BUG_ON(!ctx_obj); |
|
- | 367 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj)); |
|
- | 368 | WARN_ON(!i915_gem_obj_is_pinned(rb_obj)); |
|
367 | 369 | ||
368 | page = i915_gem_object_get_page(ctx_obj, 1); |
370 | page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); |
369 | reg_state = kmap_atomic(page); |
371 | reg_state = kmap_atomic(page); |
370 | 372 | ||
- | 373 | reg_state[CTX_RING_TAIL+1] = rq->tail; |
|
- | 374 | reg_state[CTX_RING_BUFFER_START+1] = i915_gem_obj_ggtt_offset(rb_obj); |
|
- | 375 | ||
- | 376 | if (ppgtt && !USES_FULL_48BIT_PPGTT(ppgtt->base.dev)) { |
|
- | 377 | /* True 32b PPGTT with dynamic page allocation: update PDP |
|
- | 378 | * registers and point the unallocated PDPs to scratch page. |
|
- | 379 | * PML4 is allocated during ppgtt init, so this is not needed |
|
- | 380 | * in 48-bit mode. |
|
- | 381 | */ |
|
- | 382 | ASSIGN_CTX_PDP(ppgtt, reg_state, 3); |
|
- | 383 | ASSIGN_CTX_PDP(ppgtt, reg_state, 2); |
|
- | 384 | ASSIGN_CTX_PDP(ppgtt, reg_state, 1); |
|
371 | reg_state[CTX_RING_TAIL+1] = tail; |
385 | ASSIGN_CTX_PDP(ppgtt, reg_state, 0); |
372 | reg_state[CTX_RING_BUFFER_START+1] = i915_gem_obj_ggtt_offset(ring_obj); |
386 | } |
373 | 387 | ||
374 | kunmap_atomic(reg_state); |
388 | kunmap_atomic(reg_state); |
375 | 389 | ||
376 | return 0; |
390 | return 0; |
377 | } |
391 | } |
378 | 392 | ||
379 | static void execlists_submit_contexts(struct intel_engine_cs *ring, |
- | |
380 | struct intel_context *to0, u32 tail0, |
393 | static void execlists_submit_requests(struct drm_i915_gem_request *rq0, |
381 | struct intel_context *to1, u32 tail1) |
394 | struct drm_i915_gem_request *rq1) |
382 | { |
- | |
383 | struct drm_i915_gem_object *ctx_obj0 = to0->engine[ring->id].state; |
- | |
384 | struct intel_ringbuffer *ringbuf0 = to0->engine[ring->id].ringbuf; |
- | |
385 | struct drm_i915_gem_object *ctx_obj1 = NULL; |
- | |
386 | struct intel_ringbuffer *ringbuf1 = NULL; |
- | |
387 | - | ||
388 | BUG_ON(!ctx_obj0); |
- | |
389 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj0)); |
- | |
390 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf0->obj)); |
- | |
391 | 395 | { |
|
392 | execlists_update_context(ctx_obj0, ringbuf0->obj, tail0); |
396 | execlists_update_context(rq0); |
393 | - | ||
394 | if (to1) { |
- | |
395 | ringbuf1 = to1->engine[ring->id].ringbuf; |
397 | |
396 | ctx_obj1 = to1->engine[ring->id].state; |
- | |
397 | BUG_ON(!ctx_obj1); |
- | |
398 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj1)); |
- | |
399 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf1->obj)); |
- | |
400 | - | ||
401 | execlists_update_context(ctx_obj1, ringbuf1->obj, tail1); |
398 | if (rq1) |
402 | } |
399 | execlists_update_context(rq1); |
403 | 400 | ||
404 | execlists_elsp_write(ring, ctx_obj0, ctx_obj1); |
401 | execlists_elsp_write(rq0, rq1); |
405 | } |
402 | } |
406 | 403 | ||
407 | static void execlists_context_unqueue(struct intel_engine_cs *ring) |
404 | static void execlists_context_unqueue(struct intel_engine_cs *ring) |
408 | { |
405 | { |
409 | struct intel_ctx_submit_request *req0 = NULL, *req1 = NULL; |
406 | struct drm_i915_gem_request *req0 = NULL, *req1 = NULL; |
410 | struct intel_ctx_submit_request *cursor = NULL, *tmp = NULL; |
407 | struct drm_i915_gem_request *cursor = NULL, *tmp = NULL; |
411 | 408 | ||
412 | assert_spin_locked(&ring->execlist_lock); |
409 | assert_spin_locked(&ring->execlist_lock); |
- | 410 | ||
- | 411 | /* |
|
- | 412 | * If irqs are not active generate a warning as batches that finish |
|
- | 413 | * without the irqs may get lost and a GPU Hang may occur. |
|
- | 414 | */ |
|
- | 415 | WARN_ON(!intel_irqs_enabled(ring->dev->dev_private)); |
|
413 | 416 | ||
414 | if (list_empty(&ring->execlist_queue)) |
417 | if (list_empty(&ring->execlist_queue)) |
415 | return; |
418 | return; |
416 | 419 | ||
417 | /* Try to read in pairs */ |
420 | /* Try to read in pairs */ |
418 | list_for_each_entry_safe(cursor, tmp, &ring->execlist_queue, |
421 | list_for_each_entry_safe(cursor, tmp, &ring->execlist_queue, |
419 | execlist_link) { |
422 | execlist_link) { |
420 | if (!req0) { |
423 | if (!req0) { |
421 | req0 = cursor; |
424 | req0 = cursor; |
422 | } else if (req0->ctx == cursor->ctx) { |
425 | } else if (req0->ctx == cursor->ctx) { |
423 | /* Same ctx: ignore first request, as second request |
426 | /* Same ctx: ignore first request, as second request |
424 | * will update tail past first request's workload */ |
427 | * will update tail past first request's workload */ |
425 | cursor->elsp_submitted = req0->elsp_submitted; |
428 | cursor->elsp_submitted = req0->elsp_submitted; |
426 | list_del(&req0->execlist_link); |
429 | list_del(&req0->execlist_link); |
427 | list_add_tail(&req0->execlist_link, |
430 | list_add_tail(&req0->execlist_link, |
428 | &ring->execlist_retired_req_list); |
431 | &ring->execlist_retired_req_list); |
429 | req0 = cursor; |
432 | req0 = cursor; |
430 | } else { |
433 | } else { |
431 | req1 = cursor; |
434 | req1 = cursor; |
432 | break; |
435 | break; |
433 | } |
436 | } |
434 | } |
437 | } |
- | 438 | ||
- | 439 | if (IS_GEN8(ring->dev) || IS_GEN9(ring->dev)) { |
|
- | 440 | /* |
|
- | 441 | * WaIdleLiteRestore: make sure we never cause a lite |
|
- | 442 | * restore with HEAD==TAIL |
|
- | 443 | */ |
|
- | 444 | if (req0->elsp_submitted) { |
|
- | 445 | /* |
|
- | 446 | * Apply the wa NOOPS to prevent ring:HEAD == req:TAIL |
|
- | 447 | * as we resubmit the request. See gen8_emit_request() |
|
- | 448 | * for where we prepare the padding after the end of the |
|
- | 449 | * request. |
|
- | 450 | */ |
|
- | 451 | struct intel_ringbuffer *ringbuf; |
|
- | 452 | ||
- | 453 | ringbuf = req0->ctx->engine[ring->id].ringbuf; |
|
- | 454 | req0->tail += 8; |
|
- | 455 | req0->tail &= ringbuf->size - 1; |
|
- | 456 | } |
|
- | 457 | } |
|
435 | 458 | ||
436 | WARN_ON(req1 && req1->elsp_submitted); |
459 | WARN_ON(req1 && req1->elsp_submitted); |
437 | - | ||
438 | execlists_submit_contexts(ring, req0->ctx, req0->tail, |
- | |
439 | req1 ? req1->ctx : NULL, |
- | |
440 | req1 ? req1->tail : 0); |
- | |
441 | - | ||
442 | req0->elsp_submitted++; |
- | |
443 | if (req1) |
460 | |
444 | req1->elsp_submitted++; |
461 | execlists_submit_requests(req0, req1); |
445 | } |
462 | } |
446 | 463 | ||
447 | static bool execlists_check_remove_request(struct intel_engine_cs *ring, |
464 | static bool execlists_check_remove_request(struct intel_engine_cs *ring, |
448 | u32 request_id) |
465 | u32 request_id) |
449 | { |
466 | { |
450 | struct intel_ctx_submit_request *head_req; |
467 | struct drm_i915_gem_request *head_req; |
451 | 468 | ||
452 | assert_spin_locked(&ring->execlist_lock); |
469 | assert_spin_locked(&ring->execlist_lock); |
453 | 470 | ||
454 | head_req = list_first_entry_or_null(&ring->execlist_queue, |
471 | head_req = list_first_entry_or_null(&ring->execlist_queue, |
455 | struct intel_ctx_submit_request, |
472 | struct drm_i915_gem_request, |
456 | execlist_link); |
473 | execlist_link); |
457 | 474 | ||
458 | if (head_req != NULL) { |
475 | if (head_req != NULL) { |
459 | struct drm_i915_gem_object *ctx_obj = |
476 | struct drm_i915_gem_object *ctx_obj = |
460 | head_req->ctx->engine[ring->id].state; |
477 | head_req->ctx->engine[ring->id].state; |
461 | if (intel_execlists_ctx_id(ctx_obj) == request_id) { |
478 | if (intel_execlists_ctx_id(ctx_obj) == request_id) { |
462 | WARN(head_req->elsp_submitted == 0, |
479 | WARN(head_req->elsp_submitted == 0, |
463 | "Never submitted head request\n"); |
480 | "Never submitted head request\n"); |
464 | 481 | ||
465 | if (--head_req->elsp_submitted <= 0) { |
482 | if (--head_req->elsp_submitted <= 0) { |
466 | list_del(&head_req->execlist_link); |
483 | list_del(&head_req->execlist_link); |
467 | list_add_tail(&head_req->execlist_link, |
484 | list_add_tail(&head_req->execlist_link, |
468 | &ring->execlist_retired_req_list); |
485 | &ring->execlist_retired_req_list); |
469 | return true; |
486 | return true; |
470 | } |
487 | } |
471 | } |
488 | } |
472 | } |
489 | } |
473 | 490 | ||
474 | return false; |
491 | return false; |
475 | } |
492 | } |
476 | 493 | ||
477 | /** |
494 | /** |
478 | * intel_execlists_handle_ctx_events() - handle Context Switch interrupts |
495 | * intel_lrc_irq_handler() - handle Context Switch interrupts |
479 | * @ring: Engine Command Streamer to handle. |
496 | * @ring: Engine Command Streamer to handle. |
480 | * |
497 | * |
481 | * Check the unread Context Status Buffers and manage the submission of new |
498 | * Check the unread Context Status Buffers and manage the submission of new |
482 | * contexts to the ELSP accordingly. |
499 | * contexts to the ELSP accordingly. |
483 | */ |
500 | */ |
484 | void intel_execlists_handle_ctx_events(struct intel_engine_cs *ring) |
501 | void intel_lrc_irq_handler(struct intel_engine_cs *ring) |
485 | { |
502 | { |
486 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
503 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
487 | u32 status_pointer; |
504 | u32 status_pointer; |
488 | u8 read_pointer; |
505 | u8 read_pointer; |
489 | u8 write_pointer; |
506 | u8 write_pointer; |
490 | u32 status; |
507 | u32 status = 0; |
491 | u32 status_id; |
508 | u32 status_id; |
492 | u32 submit_contexts = 0; |
509 | u32 submit_contexts = 0; |
493 | 510 | ||
494 | status_pointer = I915_READ(RING_CONTEXT_STATUS_PTR(ring)); |
511 | status_pointer = I915_READ(RING_CONTEXT_STATUS_PTR(ring)); |
495 | 512 | ||
496 | read_pointer = ring->next_context_status_buffer; |
513 | read_pointer = ring->next_context_status_buffer; |
497 | write_pointer = status_pointer & 0x07; |
514 | write_pointer = status_pointer & GEN8_CSB_PTR_MASK; |
498 | if (read_pointer > write_pointer) |
515 | if (read_pointer > write_pointer) |
499 | write_pointer += 6; |
516 | write_pointer += GEN8_CSB_ENTRIES; |
500 | 517 | ||
501 | spin_lock(&ring->execlist_lock); |
518 | spin_lock(&ring->execlist_lock); |
502 | 519 | ||
503 | while (read_pointer < write_pointer) { |
520 | while (read_pointer < write_pointer) { |
504 | read_pointer++; |
521 | read_pointer++; |
505 | status = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + |
522 | status = I915_READ(RING_CONTEXT_STATUS_BUF_LO(ring, read_pointer % GEN8_CSB_ENTRIES)); |
506 | (read_pointer % 6) * 8); |
523 | status_id = I915_READ(RING_CONTEXT_STATUS_BUF_HI(ring, read_pointer % GEN8_CSB_ENTRIES)); |
- | 524 | ||
507 | status_id = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + |
525 | if (status & GEN8_CTX_STATUS_IDLE_ACTIVE) |
508 | (read_pointer % 6) * 8 + 4); |
526 | continue; |
509 | 527 | ||
510 | if (status & GEN8_CTX_STATUS_PREEMPTED) { |
528 | if (status & GEN8_CTX_STATUS_PREEMPTED) { |
511 | if (status & GEN8_CTX_STATUS_LITE_RESTORE) { |
529 | if (status & GEN8_CTX_STATUS_LITE_RESTORE) { |
512 | if (execlists_check_remove_request(ring, status_id)) |
530 | if (execlists_check_remove_request(ring, status_id)) |
513 | WARN(1, "Lite Restored request removed from queue\n"); |
531 | WARN(1, "Lite Restored request removed from queue\n"); |
514 | } else |
532 | } else |
515 | WARN(1, "Preemption without Lite Restore\n"); |
533 | WARN(1, "Preemption without Lite Restore\n"); |
516 | } |
534 | } |
517 | 535 | ||
518 | if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) || |
536 | if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) || |
519 | (status & GEN8_CTX_STATUS_ELEMENT_SWITCH)) { |
537 | (status & GEN8_CTX_STATUS_ELEMENT_SWITCH)) { |
520 | if (execlists_check_remove_request(ring, status_id)) |
538 | if (execlists_check_remove_request(ring, status_id)) |
521 | submit_contexts++; |
539 | submit_contexts++; |
522 | } |
540 | } |
523 | } |
541 | } |
- | 542 | ||
- | 543 | if (disable_lite_restore_wa(ring)) { |
|
- | 544 | /* Prevent a ctx to preempt itself */ |
|
524 | 545 | if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) && |
|
525 | if (submit_contexts != 0) |
546 | (submit_contexts != 0)) |
- | 547 | execlists_context_unqueue(ring); |
|
- | 548 | } else if (submit_contexts != 0) { |
|
- | 549 | execlists_context_unqueue(ring); |
|
526 | execlists_context_unqueue(ring); |
550 | } |
527 | 551 | ||
528 | spin_unlock(&ring->execlist_lock); |
552 | spin_unlock(&ring->execlist_lock); |
529 | 553 | ||
530 | WARN(submit_contexts > 2, "More than two context complete events?\n"); |
554 | WARN(submit_contexts > 2, "More than two context complete events?\n"); |
531 | ring->next_context_status_buffer = write_pointer % 6; |
555 | ring->next_context_status_buffer = write_pointer % GEN8_CSB_ENTRIES; |
532 | 556 | ||
- | 557 | I915_WRITE(RING_CONTEXT_STATUS_PTR(ring), |
|
533 | I915_WRITE(RING_CONTEXT_STATUS_PTR(ring), |
558 | _MASKED_FIELD(GEN8_CSB_PTR_MASK << 8, |
- | 559 | ((u32)ring->next_context_status_buffer & |
|
534 | ((u32)ring->next_context_status_buffer & 0x07) << 8); |
560 | GEN8_CSB_PTR_MASK) << 8)); |
535 | } |
561 | } |
536 | - | ||
537 | static int execlists_context_queue(struct intel_engine_cs *ring, |
- | |
538 | struct intel_context *to, |
562 | |
539 | u32 tail) |
563 | static int execlists_context_queue(struct drm_i915_gem_request *request) |
540 | { |
564 | { |
541 | struct intel_ctx_submit_request *req = NULL, *cursor; |
- | |
542 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
565 | struct intel_engine_cs *ring = request->ring; |
543 | unsigned long flags; |
- | |
544 | int num_elements = 0; |
- | |
545 | - | ||
546 | req = kzalloc(sizeof(*req), GFP_KERNEL); |
- | |
547 | if (req == NULL) |
- | |
548 | return -ENOMEM; |
- | |
549 | req->ctx = to; |
566 | struct drm_i915_gem_request *cursor; |
550 | i915_gem_context_reference(req->ctx); |
567 | int num_elements = 0; |
551 | - | ||
552 | if (to != ring->default_context) |
- | |
553 | intel_lr_context_pin(ring, to); |
- | |
554 | 568 | ||
555 | req->ring = ring; |
569 | if (request->ctx != ring->default_context) |
556 | req->tail = tail; |
570 | intel_lr_context_pin(request); |
557 | 571 | ||
558 | intel_runtime_pm_get(dev_priv); |
572 | i915_gem_request_reference(request); |
559 | 573 | ||
560 | spin_lock_irqsave(&ring->execlist_lock, flags); |
574 | spin_lock_irq(&ring->execlist_lock); |
561 | 575 | ||
562 | list_for_each_entry(cursor, &ring->execlist_queue, execlist_link) |
576 | list_for_each_entry(cursor, &ring->execlist_queue, execlist_link) |
563 | if (++num_elements > 2) |
577 | if (++num_elements > 2) |
564 | break; |
578 | break; |
565 | 579 | ||
566 | if (num_elements > 2) { |
580 | if (num_elements > 2) { |
567 | struct intel_ctx_submit_request *tail_req; |
581 | struct drm_i915_gem_request *tail_req; |
568 | 582 | ||
569 | tail_req = list_last_entry(&ring->execlist_queue, |
583 | tail_req = list_last_entry(&ring->execlist_queue, |
570 | struct intel_ctx_submit_request, |
584 | struct drm_i915_gem_request, |
571 | execlist_link); |
585 | execlist_link); |
572 | 586 | ||
573 | if (to == tail_req->ctx) { |
587 | if (request->ctx == tail_req->ctx) { |
574 | WARN(tail_req->elsp_submitted != 0, |
588 | WARN(tail_req->elsp_submitted != 0, |
575 | "More than 2 already-submitted reqs queued\n"); |
589 | "More than 2 already-submitted reqs queued\n"); |
576 | list_del(&tail_req->execlist_link); |
590 | list_del(&tail_req->execlist_link); |
577 | list_add_tail(&tail_req->execlist_link, |
591 | list_add_tail(&tail_req->execlist_link, |
578 | &ring->execlist_retired_req_list); |
592 | &ring->execlist_retired_req_list); |
579 | } |
593 | } |
580 | } |
594 | } |
581 | 595 | ||
582 | list_add_tail(&req->execlist_link, &ring->execlist_queue); |
596 | list_add_tail(&request->execlist_link, &ring->execlist_queue); |
583 | if (num_elements == 0) |
597 | if (num_elements == 0) |
584 | execlists_context_unqueue(ring); |
598 | execlists_context_unqueue(ring); |
585 | 599 | ||
586 | spin_unlock_irqrestore(&ring->execlist_lock, flags); |
600 | spin_unlock_irq(&ring->execlist_lock); |
587 | 601 | ||
588 | return 0; |
602 | return 0; |
589 | } |
603 | } |
590 | 604 | ||
591 | static int logical_ring_invalidate_all_caches(struct intel_ringbuffer *ringbuf) |
605 | static int logical_ring_invalidate_all_caches(struct drm_i915_gem_request *req) |
592 | { |
606 | { |
593 | struct intel_engine_cs *ring = ringbuf->ring; |
607 | struct intel_engine_cs *ring = req->ring; |
594 | uint32_t flush_domains; |
608 | uint32_t flush_domains; |
595 | int ret; |
609 | int ret; |
596 | 610 | ||
597 | flush_domains = 0; |
611 | flush_domains = 0; |
598 | if (ring->gpu_caches_dirty) |
612 | if (ring->gpu_caches_dirty) |
599 | flush_domains = I915_GEM_GPU_DOMAINS; |
613 | flush_domains = I915_GEM_GPU_DOMAINS; |
600 | 614 | ||
601 | ret = ring->emit_flush(ringbuf, I915_GEM_GPU_DOMAINS, flush_domains); |
615 | ret = ring->emit_flush(req, I915_GEM_GPU_DOMAINS, flush_domains); |
602 | if (ret) |
616 | if (ret) |
603 | return ret; |
617 | return ret; |
604 | 618 | ||
605 | ring->gpu_caches_dirty = false; |
619 | ring->gpu_caches_dirty = false; |
606 | return 0; |
620 | return 0; |
607 | } |
621 | } |
608 | 622 | ||
609 | static int execlists_move_to_gpu(struct intel_ringbuffer *ringbuf, |
623 | static int execlists_move_to_gpu(struct drm_i915_gem_request *req, |
610 | struct list_head *vmas) |
624 | struct list_head *vmas) |
611 | { |
625 | { |
612 | struct intel_engine_cs *ring = ringbuf->ring; |
626 | const unsigned other_rings = ~intel_ring_flag(req->ring); |
613 | struct i915_vma *vma; |
627 | struct i915_vma *vma; |
614 | uint32_t flush_domains = 0; |
628 | uint32_t flush_domains = 0; |
615 | bool flush_chipset = false; |
629 | bool flush_chipset = false; |
616 | int ret; |
630 | int ret; |
617 | 631 | ||
618 | list_for_each_entry(vma, vmas, exec_list) { |
632 | list_for_each_entry(vma, vmas, exec_list) { |
619 | struct drm_i915_gem_object *obj = vma->obj; |
633 | struct drm_i915_gem_object *obj = vma->obj; |
- | 634 | ||
620 | 635 | if (obj->active & other_rings) { |
|
621 | ret = i915_gem_object_sync(obj, ring); |
636 | ret = i915_gem_object_sync(obj, req->ring, &req); |
622 | if (ret) |
637 | if (ret) |
- | 638 | return ret; |
|
623 | return ret; |
639 | } |
624 | 640 | ||
625 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
641 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
626 | flush_chipset |= i915_gem_clflush_object(obj, false); |
642 | flush_chipset |= i915_gem_clflush_object(obj, false); |
627 | 643 | ||
628 | flush_domains |= obj->base.write_domain; |
644 | flush_domains |= obj->base.write_domain; |
629 | } |
645 | } |
630 | 646 | ||
631 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
647 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
632 | wmb(); |
648 | wmb(); |
633 | 649 | ||
634 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
650 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
635 | * any residual writes from the previous batch. |
651 | * any residual writes from the previous batch. |
636 | */ |
652 | */ |
637 | return logical_ring_invalidate_all_caches(ringbuf); |
653 | return logical_ring_invalidate_all_caches(req); |
- | 654 | } |
|
- | 655 | ||
- | 656 | int intel_logical_ring_alloc_request_extras(struct drm_i915_gem_request *request) |
|
- | 657 | { |
|
- | 658 | int ret; |
|
- | 659 | ||
- | 660 | request->ringbuf = request->ctx->engine[request->ring->id].ringbuf; |
|
- | 661 | ||
- | 662 | if (request->ctx != request->ring->default_context) { |
|
- | 663 | ret = intel_lr_context_pin(request); |
|
- | 664 | if (ret) |
|
- | 665 | return ret; |
|
- | 666 | } |
|
- | 667 | ||
- | 668 | return 0; |
|
- | 669 | } |
|
- | 670 | ||
- | 671 | static int logical_ring_wait_for_space(struct drm_i915_gem_request *req, |
|
- | 672 | int bytes) |
|
- | 673 | { |
|
- | 674 | struct intel_ringbuffer *ringbuf = req->ringbuf; |
|
- | 675 | struct intel_engine_cs *ring = req->ring; |
|
- | 676 | struct drm_i915_gem_request *target; |
|
- | 677 | unsigned space; |
|
- | 678 | int ret; |
|
- | 679 | ||
- | 680 | if (intel_ring_space(ringbuf) >= bytes) |
|
- | 681 | return 0; |
|
- | 682 | ||
- | 683 | /* The whole point of reserving space is to not wait! */ |
|
- | 684 | WARN_ON(ringbuf->reserved_in_use); |
|
- | 685 | ||
- | 686 | list_for_each_entry(target, &ring->request_list, list) { |
|
- | 687 | /* |
|
- | 688 | * The request queue is per-engine, so can contain requests |
|
- | 689 | * from multiple ringbuffers. Here, we must ignore any that |
|
- | 690 | * aren't from the ringbuffer we're considering. |
|
- | 691 | */ |
|
- | 692 | if (target->ringbuf != ringbuf) |
|
- | 693 | continue; |
|
- | 694 | ||
- | 695 | /* Would completion of this request free enough space? */ |
|
- | 696 | space = __intel_ring_space(target->postfix, ringbuf->tail, |
|
- | 697 | ringbuf->size); |
|
- | 698 | if (space >= bytes) |
|
- | 699 | break; |
|
- | 700 | } |
|
- | 701 | ||
- | 702 | if (WARN_ON(&target->list == &ring->request_list)) |
|
- | 703 | return -ENOSPC; |
|
- | 704 | ||
- | 705 | ret = i915_wait_request(target); |
|
- | 706 | if (ret) |
|
- | 707 | return ret; |
|
- | 708 | ||
- | 709 | ringbuf->space = space; |
|
- | 710 | return 0; |
|
- | 711 | } |
|
- | 712 | ||
- | 713 | /* |
|
- | 714 | * intel_logical_ring_advance_and_submit() - advance the tail and submit the workload |
|
- | 715 | * @request: Request to advance the logical ringbuffer of. |
|
- | 716 | * |
|
- | 717 | * The tail is updated in our logical ringbuffer struct, not in the actual context. What |
|
- | 718 | * really happens during submission is that the context and current tail will be placed |
|
- | 719 | * on a queue waiting for the ELSP to be ready to accept a new context submission. At that |
|
- | 720 | * point, the tail *inside* the context is updated and the ELSP written to. |
|
- | 721 | */ |
|
- | 722 | static void |
|
- | 723 | intel_logical_ring_advance_and_submit(struct drm_i915_gem_request *request) |
|
- | 724 | { |
|
- | 725 | struct intel_engine_cs *ring = request->ring; |
|
- | 726 | struct drm_i915_private *dev_priv = request->i915; |
|
- | 727 | ||
- | 728 | intel_logical_ring_advance(request->ringbuf); |
|
- | 729 | ||
- | 730 | request->tail = request->ringbuf->tail; |
|
- | 731 | ||
- | 732 | if (intel_ring_stopped(ring)) |
|
- | 733 | return; |
|
- | 734 | ||
- | 735 | if (dev_priv->guc.execbuf_client) |
|
- | 736 | i915_guc_submit(dev_priv->guc.execbuf_client, request); |
|
- | 737 | else |
|
- | 738 | execlists_context_queue(request); |
|
- | 739 | } |
|
- | 740 | ||
- | 741 | static void __wrap_ring_buffer(struct intel_ringbuffer *ringbuf) |
|
- | 742 | { |
|
- | 743 | uint32_t __iomem *virt; |
|
- | 744 | int rem = ringbuf->size - ringbuf->tail; |
|
- | 745 | ||
- | 746 | virt = ringbuf->virtual_start + ringbuf->tail; |
|
- | 747 | rem /= 4; |
|
- | 748 | while (rem--) |
|
- | 749 | iowrite32(MI_NOOP, virt++); |
|
- | 750 | ||
- | 751 | ringbuf->tail = 0; |
|
- | 752 | intel_ring_update_space(ringbuf); |
|
- | 753 | } |
|
- | 754 | ||
- | 755 | static int logical_ring_prepare(struct drm_i915_gem_request *req, int bytes) |
|
- | 756 | { |
|
- | 757 | struct intel_ringbuffer *ringbuf = req->ringbuf; |
|
- | 758 | int remain_usable = ringbuf->effective_size - ringbuf->tail; |
|
- | 759 | int remain_actual = ringbuf->size - ringbuf->tail; |
|
- | 760 | int ret, total_bytes, wait_bytes = 0; |
|
- | 761 | bool need_wrap = false; |
|
- | 762 | ||
- | 763 | if (ringbuf->reserved_in_use) |
|
- | 764 | total_bytes = bytes; |
|
- | 765 | else |
|
- | 766 | total_bytes = bytes + ringbuf->reserved_size; |
|
- | 767 | ||
- | 768 | if (unlikely(bytes > remain_usable)) { |
|
- | 769 | /* |
|
- | 770 | * Not enough space for the basic request. So need to flush |
|
- | 771 | * out the remainder and then wait for base + reserved. |
|
- | 772 | */ |
|
- | 773 | wait_bytes = remain_actual + total_bytes; |
|
- | 774 | need_wrap = true; |
|
- | 775 | } else { |
|
- | 776 | if (unlikely(total_bytes > remain_usable)) { |
|
- | 777 | /* |
|
- | 778 | * The base request will fit but the reserved space |
|
- | 779 | * falls off the end. So only need to to wait for the |
|
- | 780 | * reserved size after flushing out the remainder. |
|
- | 781 | */ |
|
- | 782 | wait_bytes = remain_actual + ringbuf->reserved_size; |
|
- | 783 | need_wrap = true; |
|
- | 784 | } else if (total_bytes > ringbuf->space) { |
|
- | 785 | /* No wrapping required, just waiting. */ |
|
- | 786 | wait_bytes = total_bytes; |
|
- | 787 | } |
|
- | 788 | } |
|
- | 789 | ||
- | 790 | if (wait_bytes) { |
|
- | 791 | ret = logical_ring_wait_for_space(req, wait_bytes); |
|
- | 792 | if (unlikely(ret)) |
|
- | 793 | return ret; |
|
- | 794 | ||
- | 795 | if (need_wrap) |
|
- | 796 | __wrap_ring_buffer(ringbuf); |
|
- | 797 | } |
|
- | 798 | ||
- | 799 | return 0; |
|
- | 800 | } |
|
- | 801 | ||
- | 802 | /** |
|
- | 803 | * intel_logical_ring_begin() - prepare the logical ringbuffer to accept some commands |
|
- | 804 | * |
|
- | 805 | * @req: The request to start some new work for |
|
- | 806 | * @num_dwords: number of DWORDs that we plan to write to the ringbuffer. |
|
- | 807 | * |
|
- | 808 | * The ringbuffer might not be ready to accept the commands right away (maybe it needs to |
|
- | 809 | * be wrapped, or wait a bit for the tail to be updated). This function takes care of that |
|
- | 810 | * and also preallocates a request (every workload submission is still mediated through |
|
- | 811 | * requests, same as it did with legacy ringbuffer submission). |
|
- | 812 | * |
|
- | 813 | * Return: non-zero if the ringbuffer is not ready to be written to. |
|
- | 814 | */ |
|
- | 815 | int intel_logical_ring_begin(struct drm_i915_gem_request *req, int num_dwords) |
|
- | 816 | { |
|
- | 817 | struct drm_i915_private *dev_priv; |
|
- | 818 | int ret; |
|
- | 819 | ||
- | 820 | WARN_ON(req == NULL); |
|
- | 821 | dev_priv = req->ring->dev->dev_private; |
|
- | 822 | ||
- | 823 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, |
|
- | 824 | dev_priv->mm.interruptible); |
|
- | 825 | if (ret) |
|
- | 826 | return ret; |
|
- | 827 | ||
- | 828 | ret = logical_ring_prepare(req, num_dwords * sizeof(uint32_t)); |
|
- | 829 | if (ret) |
|
- | 830 | return ret; |
|
- | 831 | ||
- | 832 | req->ringbuf->space -= num_dwords * sizeof(uint32_t); |
|
- | 833 | return 0; |
|
- | 834 | } |
|
- | 835 | ||
- | 836 | int intel_logical_ring_reserve_space(struct drm_i915_gem_request *request) |
|
- | 837 | { |
|
- | 838 | /* |
|
- | 839 | * The first call merely notes the reserve request and is common for |
|
- | 840 | * all back ends. The subsequent localised _begin() call actually |
|
- | 841 | * ensures that the reservation is available. Without the begin, if |
|
- | 842 | * the request creator immediately submitted the request without |
|
- | 843 | * adding any commands to it then there might not actually be |
|
- | 844 | * sufficient room for the submission commands. |
|
- | 845 | */ |
|
- | 846 | intel_ring_reserved_space_reserve(request->ringbuf, MIN_SPACE_FOR_ADD_REQUEST); |
|
- | 847 | ||
- | 848 | return intel_logical_ring_begin(request, 0); |
|
638 | } |
849 | } |
639 | 850 | ||
640 | /** |
851 | /** |
641 | * execlists_submission() - submit a batchbuffer for execution, Execlists style |
852 | * execlists_submission() - submit a batchbuffer for execution, Execlists style |
642 | * @dev: DRM device. |
853 | * @dev: DRM device. |
643 | * @file: DRM file. |
854 | * @file: DRM file. |
644 | * @ring: Engine Command Streamer to submit to. |
855 | * @ring: Engine Command Streamer to submit to. |
645 | * @ctx: Context to employ for this submission. |
856 | * @ctx: Context to employ for this submission. |
646 | * @args: execbuffer call arguments. |
857 | * @args: execbuffer call arguments. |
647 | * @vmas: list of vmas. |
858 | * @vmas: list of vmas. |
648 | * @batch_obj: the batchbuffer to submit. |
859 | * @batch_obj: the batchbuffer to submit. |
649 | * @exec_start: batchbuffer start virtual address pointer. |
860 | * @exec_start: batchbuffer start virtual address pointer. |
650 | * @flags: translated execbuffer call flags. |
861 | * @dispatch_flags: translated execbuffer call flags. |
651 | * |
862 | * |
652 | * This is the evil twin version of i915_gem_ringbuffer_submission. It abstracts |
863 | * This is the evil twin version of i915_gem_ringbuffer_submission. It abstracts |
653 | * away the submission details of the execbuffer ioctl call. |
864 | * away the submission details of the execbuffer ioctl call. |
654 | * |
865 | * |
655 | * Return: non-zero if the submission fails. |
866 | * Return: non-zero if the submission fails. |
656 | */ |
867 | */ |
657 | int intel_execlists_submission(struct drm_device *dev, struct drm_file *file, |
868 | int intel_execlists_submission(struct i915_execbuffer_params *params, |
658 | struct intel_engine_cs *ring, |
- | |
659 | struct intel_context *ctx, |
- | |
660 | struct drm_i915_gem_execbuffer2 *args, |
869 | struct drm_i915_gem_execbuffer2 *args, |
661 | struct list_head *vmas, |
870 | struct list_head *vmas) |
662 | struct drm_i915_gem_object *batch_obj, |
- | |
663 | u64 exec_start, u32 flags) |
- | |
664 | { |
871 | { |
- | 872 | struct drm_device *dev = params->dev; |
|
- | 873 | struct intel_engine_cs *ring = params->ring; |
|
665 | struct drm_i915_private *dev_priv = dev->dev_private; |
874 | struct drm_i915_private *dev_priv = dev->dev_private; |
666 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
875 | struct intel_ringbuffer *ringbuf = params->ctx->engine[ring->id].ringbuf; |
- | 876 | u64 exec_start; |
|
667 | int instp_mode; |
877 | int instp_mode; |
668 | u32 instp_mask; |
878 | u32 instp_mask; |
669 | int ret; |
879 | int ret; |
670 | 880 | ||
671 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
881 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
672 | instp_mask = I915_EXEC_CONSTANTS_MASK; |
882 | instp_mask = I915_EXEC_CONSTANTS_MASK; |
673 | switch (instp_mode) { |
883 | switch (instp_mode) { |
674 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
884 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
675 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
885 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
676 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
886 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
677 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { |
887 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { |
678 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); |
888 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); |
679 | return -EINVAL; |
889 | return -EINVAL; |
680 | } |
890 | } |
681 | 891 | ||
682 | if (instp_mode != dev_priv->relative_constants_mode) { |
892 | if (instp_mode != dev_priv->relative_constants_mode) { |
683 | if (instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { |
893 | if (instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { |
684 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); |
894 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); |
685 | return -EINVAL; |
895 | return -EINVAL; |
686 | } |
896 | } |
687 | 897 | ||
688 | /* The HW changed the meaning on this bit on gen6 */ |
898 | /* The HW changed the meaning on this bit on gen6 */ |
689 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
899 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
690 | } |
900 | } |
691 | break; |
901 | break; |
692 | default: |
902 | default: |
693 | DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); |
903 | DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); |
694 | return -EINVAL; |
904 | return -EINVAL; |
695 | } |
905 | } |
696 | - | ||
697 | if (args->num_cliprects != 0) { |
- | |
698 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); |
- | |
699 | return -EINVAL; |
- | |
700 | } else { |
- | |
701 | if (args->DR4 == 0xffffffff) { |
- | |
702 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); |
- | |
703 | args->DR4 = 0; |
- | |
704 | } |
- | |
705 | - | ||
706 | if (args->DR1 || args->DR4 || args->cliprects_ptr) { |
- | |
707 | DRM_DEBUG("0 cliprects but dirt in cliprects fields\n"); |
- | |
708 | return -EINVAL; |
- | |
709 | } |
- | |
710 | } |
- | |
711 | 906 | ||
712 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
907 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
713 | DRM_DEBUG("sol reset is gen7 only\n"); |
908 | DRM_DEBUG("sol reset is gen7 only\n"); |
714 | return -EINVAL; |
909 | return -EINVAL; |
715 | } |
910 | } |
716 | 911 | ||
717 | ret = execlists_move_to_gpu(ringbuf, vmas); |
912 | ret = execlists_move_to_gpu(params->request, vmas); |
718 | if (ret) |
913 | if (ret) |
719 | return ret; |
914 | return ret; |
720 | 915 | ||
721 | if (ring == &dev_priv->ring[RCS] && |
916 | if (ring == &dev_priv->ring[RCS] && |
722 | instp_mode != dev_priv->relative_constants_mode) { |
917 | instp_mode != dev_priv->relative_constants_mode) { |
723 | ret = intel_logical_ring_begin(ringbuf, 4); |
918 | ret = intel_logical_ring_begin(params->request, 4); |
724 | if (ret) |
919 | if (ret) |
725 | return ret; |
920 | return ret; |
726 | 921 | ||
727 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
922 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
728 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(1)); |
923 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(1)); |
729 | intel_logical_ring_emit(ringbuf, INSTPM); |
924 | intel_logical_ring_emit(ringbuf, INSTPM); |
730 | intel_logical_ring_emit(ringbuf, instp_mask << 16 | instp_mode); |
925 | intel_logical_ring_emit(ringbuf, instp_mask << 16 | instp_mode); |
731 | intel_logical_ring_advance(ringbuf); |
926 | intel_logical_ring_advance(ringbuf); |
732 | 927 | ||
733 | dev_priv->relative_constants_mode = instp_mode; |
928 | dev_priv->relative_constants_mode = instp_mode; |
734 | } |
929 | } |
- | 930 | ||
- | 931 | exec_start = params->batch_obj_vm_offset + |
|
- | 932 | args->batch_start_offset; |
|
735 | 933 | ||
736 | ret = ring->emit_bb_start(ringbuf, exec_start, flags); |
934 | ret = ring->emit_bb_start(params->request, exec_start, params->dispatch_flags); |
737 | if (ret) |
935 | if (ret) |
738 | return ret; |
936 | return ret; |
- | 937 | ||
- | 938 | trace_i915_gem_ring_dispatch(params->request, params->dispatch_flags); |
|
739 | 939 | ||
740 | i915_gem_execbuffer_move_to_active(vmas, ring); |
940 | i915_gem_execbuffer_move_to_active(vmas, params->request); |
741 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); |
941 | i915_gem_execbuffer_retire_commands(params); |
742 | 942 | ||
743 | return 0; |
943 | return 0; |
744 | } |
944 | } |
745 | 945 | ||
746 | void intel_execlists_retire_requests(struct intel_engine_cs *ring) |
946 | void intel_execlists_retire_requests(struct intel_engine_cs *ring) |
747 | { |
947 | { |
748 | struct intel_ctx_submit_request *req, *tmp; |
948 | struct drm_i915_gem_request *req, *tmp; |
749 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
- | |
750 | unsigned long flags; |
- | |
751 | struct list_head retired_list; |
949 | struct list_head retired_list; |
752 | 950 | ||
753 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
951 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
754 | if (list_empty(&ring->execlist_retired_req_list)) |
952 | if (list_empty(&ring->execlist_retired_req_list)) |
755 | return; |
953 | return; |
756 | 954 | ||
757 | INIT_LIST_HEAD(&retired_list); |
955 | INIT_LIST_HEAD(&retired_list); |
758 | spin_lock_irqsave(&ring->execlist_lock, flags); |
956 | spin_lock_irq(&ring->execlist_lock); |
759 | list_replace_init(&ring->execlist_retired_req_list, &retired_list); |
957 | list_replace_init(&ring->execlist_retired_req_list, &retired_list); |
760 | spin_unlock_irqrestore(&ring->execlist_lock, flags); |
958 | spin_unlock_irq(&ring->execlist_lock); |
761 | 959 | ||
762 | list_for_each_entry_safe(req, tmp, &retired_list, execlist_link) { |
960 | list_for_each_entry_safe(req, tmp, &retired_list, execlist_link) { |
763 | struct intel_context *ctx = req->ctx; |
961 | struct intel_context *ctx = req->ctx; |
764 | struct drm_i915_gem_object *ctx_obj = |
962 | struct drm_i915_gem_object *ctx_obj = |
765 | ctx->engine[ring->id].state; |
963 | ctx->engine[ring->id].state; |
766 | 964 | ||
767 | if (ctx_obj && (ctx != ring->default_context)) |
965 | if (ctx_obj && (ctx != ring->default_context)) |
768 | intel_lr_context_unpin(ring, ctx); |
966 | intel_lr_context_unpin(req); |
769 | intel_runtime_pm_put(dev_priv); |
- | |
770 | i915_gem_context_unreference(req->ctx); |
- | |
771 | list_del(&req->execlist_link); |
967 | list_del(&req->execlist_link); |
772 | kfree(req); |
968 | i915_gem_request_unreference(req); |
773 | } |
969 | } |
774 | } |
970 | } |
775 | 971 | ||
776 | void intel_logical_ring_stop(struct intel_engine_cs *ring) |
972 | void intel_logical_ring_stop(struct intel_engine_cs *ring) |
777 | { |
973 | { |
778 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
974 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
779 | int ret; |
975 | int ret; |
780 | 976 | ||
781 | if (!intel_ring_initialized(ring)) |
977 | if (!intel_ring_initialized(ring)) |
782 | return; |
978 | return; |
783 | 979 | ||
784 | ret = intel_ring_idle(ring); |
980 | ret = intel_ring_idle(ring); |
785 | if (ret && !i915_reset_in_progress(&to_i915(ring->dev)->gpu_error)) |
981 | if (ret && !i915_reset_in_progress(&to_i915(ring->dev)->gpu_error)) |
786 | DRM_ERROR("failed to quiesce %s whilst cleaning up: %d\n", |
982 | DRM_ERROR("failed to quiesce %s whilst cleaning up: %d\n", |
787 | ring->name, ret); |
983 | ring->name, ret); |
788 | 984 | ||
789 | /* TODO: Is this correct with Execlists enabled? */ |
985 | /* TODO: Is this correct with Execlists enabled? */ |
790 | I915_WRITE_MODE(ring, _MASKED_BIT_ENABLE(STOP_RING)); |
986 | I915_WRITE_MODE(ring, _MASKED_BIT_ENABLE(STOP_RING)); |
791 | if (wait_for_atomic((I915_READ_MODE(ring) & MODE_IDLE) != 0, 1000)) { |
987 | if (wait_for_atomic((I915_READ_MODE(ring) & MODE_IDLE) != 0, 1000)) { |
792 | DRM_ERROR("%s :timed out trying to stop ring\n", ring->name); |
988 | DRM_ERROR("%s :timed out trying to stop ring\n", ring->name); |
793 | return; |
989 | return; |
794 | } |
990 | } |
795 | I915_WRITE_MODE(ring, _MASKED_BIT_DISABLE(STOP_RING)); |
991 | I915_WRITE_MODE(ring, _MASKED_BIT_DISABLE(STOP_RING)); |
796 | } |
992 | } |
797 | 993 | ||
798 | int logical_ring_flush_all_caches(struct intel_ringbuffer *ringbuf) |
994 | int logical_ring_flush_all_caches(struct drm_i915_gem_request *req) |
799 | { |
995 | { |
800 | struct intel_engine_cs *ring = ringbuf->ring; |
996 | struct intel_engine_cs *ring = req->ring; |
801 | int ret; |
997 | int ret; |
802 | 998 | ||
803 | if (!ring->gpu_caches_dirty) |
999 | if (!ring->gpu_caches_dirty) |
804 | return 0; |
1000 | return 0; |
805 | 1001 | ||
806 | ret = ring->emit_flush(ringbuf, 0, I915_GEM_GPU_DOMAINS); |
1002 | ret = ring->emit_flush(req, 0, I915_GEM_GPU_DOMAINS); |
807 | if (ret) |
1003 | if (ret) |
808 | return ret; |
1004 | return ret; |
809 | 1005 | ||
810 | ring->gpu_caches_dirty = false; |
1006 | ring->gpu_caches_dirty = false; |
811 | return 0; |
1007 | return 0; |
812 | } |
1008 | } |
813 | - | ||
814 | /** |
- | |
815 | * intel_logical_ring_advance_and_submit() - advance the tail and submit the workload |
- | |
816 | * @ringbuf: Logical Ringbuffer to advance. |
- | |
817 | * |
- | |
818 | * The tail is updated in our logical ringbuffer struct, not in the actual context. What |
- | |
819 | * really happens during submission is that the context and current tail will be placed |
- | |
820 | * on a queue waiting for the ELSP to be ready to accept a new context submission. At that |
- | |
821 | * point, the tail *inside* the context is updated and the ELSP written to. |
- | |
822 | */ |
1009 | |
823 | void intel_logical_ring_advance_and_submit(struct intel_ringbuffer *ringbuf) |
- | |
824 | { |
1010 | static int intel_lr_context_do_pin(struct intel_engine_cs *ring, |
825 | struct intel_engine_cs *ring = ringbuf->ring; |
- | |
826 | struct intel_context *ctx = ringbuf->FIXME_lrc_ctx; |
- | |
827 | - | ||
828 | intel_logical_ring_advance(ringbuf); |
- | |
829 | - | ||
830 | if (intel_ring_stopped(ring)) |
- | |
831 | return; |
- | |
832 | - | ||
833 | execlists_context_queue(ring, ctx, ringbuf->tail); |
- | |
834 | } |
- | |
835 | - | ||
836 | static int intel_lr_context_pin(struct intel_engine_cs *ring, |
1011 | struct drm_i915_gem_object *ctx_obj, |
837 | struct intel_context *ctx) |
1012 | struct intel_ringbuffer *ringbuf) |
838 | { |
1013 | { |
839 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; |
1014 | struct drm_device *dev = ring->dev; |
840 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
1015 | struct drm_i915_private *dev_priv = dev->dev_private; |
841 | int ret = 0; |
1016 | int ret = 0; |
842 | 1017 | ||
843 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
1018 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
844 | if (ctx->engine[ring->id].unpin_count++ == 0) { |
- | |
845 | ret = i915_gem_obj_ggtt_pin(ctx_obj, |
1019 | ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, |
846 | GEN8_LR_CONTEXT_ALIGN, 0); |
1020 | PIN_OFFSET_BIAS | GUC_WOPCM_TOP); |
847 | if (ret) |
1021 | if (ret) |
848 | goto reset_unpin_count; |
1022 | return ret; |
849 | 1023 | ||
850 | ret = intel_pin_and_map_ringbuffer_obj(ring->dev, ringbuf); |
1024 | ret = intel_pin_and_map_ringbuffer_obj(ring->dev, ringbuf); |
851 | if (ret) |
1025 | if (ret) |
852 | goto unpin_ctx_obj; |
1026 | goto unpin_ctx_obj; |
- | 1027 | ||
- | 1028 | ctx_obj->dirty = true; |
|
853 | } |
1029 | |
- | 1030 | /* Invalidate GuC TLB. */ |
|
- | 1031 | if (i915.enable_guc_submission) |
|
- | 1032 | I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); |
|
854 | 1033 | ||
855 | return ret; |
1034 | return ret; |
856 | 1035 | ||
857 | unpin_ctx_obj: |
1036 | unpin_ctx_obj: |
858 | i915_gem_object_ggtt_unpin(ctx_obj); |
1037 | i915_gem_object_ggtt_unpin(ctx_obj); |
859 | reset_unpin_count: |
- | |
860 | ctx->engine[ring->id].unpin_count = 0; |
- | |
861 | 1038 | ||
862 | return ret; |
1039 | return ret; |
863 | } |
1040 | } |
864 | 1041 | ||
865 | void intel_lr_context_unpin(struct intel_engine_cs *ring, |
- | |
866 | struct intel_context *ctx) |
1042 | static int intel_lr_context_pin(struct drm_i915_gem_request *rq) |
- | 1043 | { |
|
- | 1044 | int ret = 0; |
|
867 | { |
1045 | struct intel_engine_cs *ring = rq->ring; |
868 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; |
1046 | struct drm_i915_gem_object *ctx_obj = rq->ctx->engine[ring->id].state; |
- | 1047 | struct intel_ringbuffer *ringbuf = rq->ringbuf; |
|
- | 1048 | ||
- | 1049 | if (rq->ctx->engine[ring->id].pin_count++ == 0) { |
|
- | 1050 | ret = intel_lr_context_do_pin(ring, ctx_obj, ringbuf); |
|
- | 1051 | if (ret) |
|
- | 1052 | goto reset_pin_count; |
|
- | 1053 | } |
|
- | 1054 | return ret; |
|
- | 1055 | ||
- | 1056 | reset_pin_count: |
|
- | 1057 | rq->ctx->engine[ring->id].pin_count = 0; |
|
- | 1058 | return ret; |
|
- | 1059 | } |
|
- | 1060 | ||
- | 1061 | void intel_lr_context_unpin(struct drm_i915_gem_request *rq) |
|
- | 1062 | { |
|
- | 1063 | struct intel_engine_cs *ring = rq->ring; |
|
- | 1064 | struct drm_i915_gem_object *ctx_obj = rq->ctx->engine[ring->id].state; |
|
869 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
1065 | struct intel_ringbuffer *ringbuf = rq->ringbuf; |
870 | 1066 | ||
871 | if (ctx_obj) { |
1067 | if (ctx_obj) { |
872 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
1068 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
873 | if (--ctx->engine[ring->id].unpin_count == 0) { |
1069 | if (--rq->ctx->engine[ring->id].pin_count == 0) { |
874 | intel_unpin_ringbuffer_obj(ringbuf); |
1070 | intel_unpin_ringbuffer_obj(ringbuf); |
875 | i915_gem_object_ggtt_unpin(ctx_obj); |
1071 | i915_gem_object_ggtt_unpin(ctx_obj); |
876 | } |
1072 | } |
877 | } |
1073 | } |
878 | } |
1074 | } |
879 | 1075 | ||
880 | static int logical_ring_alloc_seqno(struct intel_engine_cs *ring, |
- | |
881 | struct intel_context *ctx) |
1076 | static int intel_logical_ring_workarounds_emit(struct drm_i915_gem_request *req) |
882 | { |
1077 | { |
- | 1078 | int ret, i; |
|
- | 1079 | struct intel_engine_cs *ring = req->ring; |
|
- | 1080 | struct intel_ringbuffer *ringbuf = req->ringbuf; |
|
- | 1081 | struct drm_device *dev = ring->dev; |
|
- | 1082 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
883 | int ret; |
1083 | struct i915_workarounds *w = &dev_priv->workarounds; |
884 | 1084 | ||
885 | if (ring->outstanding_lazy_seqno) |
1085 | if (WARN_ON_ONCE(w->count == 0)) |
886 | return 0; |
1086 | return 0; |
887 | - | ||
888 | if (ring->preallocated_lazy_request == NULL) { |
- | |
889 | struct drm_i915_gem_request *request; |
1087 | |
890 | 1088 | ring->gpu_caches_dirty = true; |
|
891 | request = kmalloc(sizeof(*request), GFP_KERNEL); |
- | |
892 | if (request == NULL) |
1089 | ret = logical_ring_flush_all_caches(req); |
893 | return -ENOMEM; |
1090 | if (ret) |
894 | - | ||
895 | if (ctx != ring->default_context) { |
1091 | return ret; |
- | 1092 | ||
- | 1093 | ret = intel_logical_ring_begin(req, w->count * 2 + 2); |
|
- | 1094 | if (ret) |
|
- | 1095 | return ret; |
|
- | 1096 | ||
896 | ret = intel_lr_context_pin(ring, ctx); |
1097 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(w->count)); |
897 | if (ret) { |
- | |
- | 1098 | for (i = 0; i < w->count; i++) { |
|
898 | kfree(request); |
- | |
899 | return ret; |
- | |
900 | } |
- | |
901 | } |
- | |
902 | - | ||
903 | /* Hold a reference to the context this request belongs to |
1099 | intel_logical_ring_emit(ringbuf, w->reg[i].addr); |
904 | * (we will need it when the time comes to emit/retire the |
1100 | intel_logical_ring_emit(ringbuf, w->reg[i].value); |
- | 1101 | } |
|
905 | * request). |
1102 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
- | 1103 | ||
906 | */ |
1104 | intel_logical_ring_advance(ringbuf); |
907 | request->ctx = ctx; |
1105 | |
908 | i915_gem_context_reference(request->ctx); |
1106 | ring->gpu_caches_dirty = true; |
909 | 1107 | ret = logical_ring_flush_all_caches(req); |
|
910 | ring->preallocated_lazy_request = request; |
- | |
911 | } |
1108 | if (ret) |
- | 1109 | return ret; |
|
912 | 1110 | ||
913 | return i915_gem_get_seqno(ring->dev, &ring->outstanding_lazy_seqno); |
1111 | return 0; |
- | 1112 | } |
|
914 | } |
1113 | |
915 | - | ||
916 | static int logical_ring_wait_request(struct intel_ringbuffer *ringbuf, |
- | |
917 | int bytes) |
- | |
- | 1114 | #define wa_ctx_emit(batch, index, cmd) \ |
|
- | 1115 | do { \ |
|
- | 1116 | int __index = (index)++; \ |
|
- | 1117 | if (WARN_ON(__index >= (PAGE_SIZE / sizeof(uint32_t)))) { \ |
|
- | 1118 | return -ENOSPC; \ |
|
- | 1119 | } \ |
|
- | 1120 | batch[__index] = (cmd); \ |
|
- | 1121 | } while (0) |
|
- | 1122 | ||
- | 1123 | ||
- | 1124 | /* |
|
- | 1125 | * In this WA we need to set GEN8_L3SQCREG4[21:21] and reset it after |
|
- | 1126 | * PIPE_CONTROL instruction. This is required for the flush to happen correctly |
|
918 | { |
1127 | * but there is a slight complication as this is applied in WA batch where the |
- | 1128 | * values are only initialized once so we cannot take register value at the |
|
- | 1129 | * beginning and reuse it further; hence we save its value to memory, upload a |
|
- | 1130 | * constant value with bit21 set and then we restore it back with the saved value. |
|
919 | struct intel_engine_cs *ring = ringbuf->ring; |
1131 | * To simplify the WA, a constant value is formed by using the default value |
920 | struct drm_i915_gem_request *request; |
1132 | * of this register. This shouldn't be a problem because we are only modifying |
- | 1133 | * it for a short period and this batch in non-premptible. We can ofcourse |
|
- | 1134 | * use additional instructions that read the actual value of the register |
|
- | 1135 | * at that time and set our bit of interest but it makes the WA complicated. |
|
921 | u32 seqno = 0; |
1136 | * |
- | 1137 | * This WA is also required for Gen9 so extracting as a function avoids |
|
- | 1138 | * code duplication. |
|
- | 1139 | */ |
|
- | 1140 | static inline int gen8_emit_flush_coherentl3_wa(struct intel_engine_cs *ring, |
|
- | 1141 | uint32_t *const batch, |
|
- | 1142 | uint32_t index) |
|
- | 1143 | { |
|
922 | int ret; |
1144 | uint32_t l3sqc4_flush = (0x40400000 | GEN8_LQSC_FLUSH_COHERENT_LINES); |
923 | 1145 | ||
924 | if (ringbuf->last_retired_head != -1) { |
1146 | /* |
925 | ringbuf->head = ringbuf->last_retired_head; |
1147 | * WaDisableLSQCROPERFforOCL:skl |
926 | ringbuf->last_retired_head = -1; |
1148 | * This WA is implemented in skl_init_clock_gating() but since |
927 | - | ||
928 | ringbuf->space = intel_ring_space(ringbuf); |
- | |
- | 1149 | * this batch updates GEN8_L3SQCREG4 with default value we need to |
|
929 | if (ringbuf->space >= bytes) |
1150 | * set this bit here to retain the WA during flush. |
930 | return 0; |
1151 | */ |
- | 1152 | if (IS_SKYLAKE(ring->dev) && INTEL_REVID(ring->dev) <= SKL_REVID_E0) |
|
- | 1153 | l3sqc4_flush |= GEN8_LQSC_RO_PERF_DIS; |
|
- | 1154 | ||
931 | } |
1155 | wa_ctx_emit(batch, index, (MI_STORE_REGISTER_MEM_GEN8 | |
932 | 1156 | MI_SRM_LRM_GLOBAL_GTT)); |
|
- | 1157 | wa_ctx_emit(batch, index, GEN8_L3SQCREG4); |
|
933 | list_for_each_entry(request, &ring->request_list, list) { |
1158 | wa_ctx_emit(batch, index, ring->scratch.gtt_offset + 256); |
- | 1159 | wa_ctx_emit(batch, index, 0); |
|
- | 1160 | ||
934 | if (__intel_ring_space(request->tail, ringbuf->tail, |
1161 | wa_ctx_emit(batch, index, MI_LOAD_REGISTER_IMM(1)); |
935 | ringbuf->size) >= bytes) { |
1162 | wa_ctx_emit(batch, index, GEN8_L3SQCREG4); |
936 | seqno = request->seqno; |
1163 | wa_ctx_emit(batch, index, l3sqc4_flush); |
937 | break; |
- | |
938 | } |
1164 | |
939 | } |
1165 | wa_ctx_emit(batch, index, GFX_OP_PIPE_CONTROL(6)); |
940 | 1166 | wa_ctx_emit(batch, index, (PIPE_CONTROL_CS_STALL | |
|
941 | if (seqno == 0) |
1167 | PIPE_CONTROL_DC_FLUSH_ENABLE)); |
- | 1168 | wa_ctx_emit(batch, index, 0); |
|
942 | return -ENOSPC; |
1169 | wa_ctx_emit(batch, index, 0); |
943 | - | ||
944 | ret = i915_wait_seqno(ring, seqno); |
- | |
945 | if (ret) |
1170 | wa_ctx_emit(batch, index, 0); |
946 | return ret; |
- | |
947 | 1171 | wa_ctx_emit(batch, index, 0); |
|
948 | i915_gem_retire_requests_ring(ring); |
1172 | |
949 | ringbuf->head = ringbuf->last_retired_head; |
1173 | wa_ctx_emit(batch, index, (MI_LOAD_REGISTER_MEM_GEN8 | |
950 | ringbuf->last_retired_head = -1; |
1174 | MI_SRM_LRM_GLOBAL_GTT)); |
- | 1175 | wa_ctx_emit(batch, index, GEN8_L3SQCREG4); |
|
- | 1176 | wa_ctx_emit(batch, index, ring->scratch.gtt_offset + 256); |
|
- | 1177 | wa_ctx_emit(batch, index, 0); |
|
- | 1178 | ||
- | 1179 | return index; |
|
- | 1180 | } |
|
- | 1181 | ||
- | 1182 | static inline uint32_t wa_ctx_start(struct i915_wa_ctx_bb *wa_ctx, |
|
- | 1183 | uint32_t offset, |
|
- | 1184 | uint32_t start_alignment) |
|
- | 1185 | { |
|
- | 1186 | return wa_ctx->offset = ALIGN(offset, start_alignment); |
|
- | 1187 | } |
|
- | 1188 | ||
- | 1189 | static inline int wa_ctx_end(struct i915_wa_ctx_bb *wa_ctx, |
|
- | 1190 | uint32_t offset, |
|
- | 1191 | uint32_t size_alignment) |
|
951 | 1192 | { |
|
952 | ringbuf->space = intel_ring_space(ringbuf); |
1193 | wa_ctx->size = offset - wa_ctx->offset; |
- | 1194 | ||
- | 1195 | WARN(wa_ctx->size % size_alignment, |
|
- | 1196 | "wa_ctx_bb failed sanity checks: size %d is not aligned to %d\n", |
|
- | 1197 | wa_ctx->size, size_alignment); |
|
- | 1198 | return 0; |
|
- | 1199 | } |
|
- | 1200 | ||
- | 1201 | /** |
|
- | 1202 | * gen8_init_indirectctx_bb() - initialize indirect ctx batch with WA |
|
- | 1203 | * |
|
- | 1204 | * @ring: only applicable for RCS |
|
- | 1205 | * @wa_ctx: structure representing wa_ctx |
|
- | 1206 | * offset: specifies start of the batch, should be cache-aligned. This is updated |
|
- | 1207 | * with the offset value received as input. |
|
- | 1208 | * size: size of the batch in DWORDS but HW expects in terms of cachelines |
|
- | 1209 | * @batch: page in which WA are loaded |
|
953 | return 0; |
1210 | * @offset: This field specifies the start of the batch, it should be |
- | 1211 | * cache-aligned otherwise it is adjusted accordingly. |
|
- | 1212 | * Typically we only have one indirect_ctx and per_ctx batch buffer which are |
|
- | 1213 | * initialized at the beginning and shared across all contexts but this field |
|
- | 1214 | * helps us to have multiple batches at different offsets and select them based |
|
- | 1215 | * on a criteria. At the moment this batch always start at the beginning of the page |
|
954 | } |
1216 | * and at this point we don't have multiple wa_ctx batch buffers. |
- | 1217 | * |
|
- | 1218 | * The number of WA applied are not known at the beginning; we use this field |
|
- | 1219 | * to return the no of DWORDS written. |
|
- | 1220 | * |
|
955 | 1221 | * It is to be noted that this batch does not contain MI_BATCH_BUFFER_END |
|
- | 1222 | * so it adds NOOPs as padding to make it cacheline aligned. |
|
956 | static int logical_ring_wait_for_space(struct intel_ringbuffer *ringbuf, |
1223 | * MI_BATCH_BUFFER_END will be added to perctx batch and both of them together |
- | 1224 | * makes a complete batch buffer. |
|
- | 1225 | * |
|
- | 1226 | * Return: non-zero if we exceed the PAGE_SIZE limit. |
|
- | 1227 | */ |
|
- | 1228 | ||
- | 1229 | static int gen8_init_indirectctx_bb(struct intel_engine_cs *ring, |
|
957 | int bytes) |
1230 | struct i915_wa_ctx_bb *wa_ctx, |
- | 1231 | uint32_t *const batch, |
|
- | 1232 | uint32_t *offset) |
|
- | 1233 | { |
|
- | 1234 | uint32_t scratch_addr; |
|
- | 1235 | uint32_t index = wa_ctx_start(wa_ctx, *offset, CACHELINE_DWORDS); |
|
- | 1236 | ||
- | 1237 | /* WaDisableCtxRestoreArbitration:bdw,chv */ |
|
- | 1238 | wa_ctx_emit(batch, index, MI_ARB_ON_OFF | MI_ARB_DISABLE); |
|
- | 1239 | ||
- | 1240 | /* WaFlushCoherentL3CacheLinesAtContextSwitch:bdw */ |
|
- | 1241 | if (IS_BROADWELL(ring->dev)) { |
|
- | 1242 | int rc = gen8_emit_flush_coherentl3_wa(ring, batch, index); |
|
958 | { |
1243 | if (rc < 0) |
- | 1244 | return rc; |
|
- | 1245 | index = rc; |
|
- | 1246 | } |
|
- | 1247 | ||
- | 1248 | /* WaClearSlmSpaceAtContextSwitch:bdw,chv */ |
|
- | 1249 | /* Actual scratch location is at 128 bytes offset */ |
|
- | 1250 | scratch_addr = ring->scratch.gtt_offset + 2*CACHELINE_BYTES; |
|
- | 1251 | ||
959 | struct intel_engine_cs *ring = ringbuf->ring; |
- | |
960 | struct drm_device *dev = ring->dev; |
- | |
961 | struct drm_i915_private *dev_priv = dev->dev_private; |
1252 | wa_ctx_emit(batch, index, GFX_OP_PIPE_CONTROL(6)); |
962 | unsigned long end; |
- | |
963 | int ret; |
- | |
964 | - | ||
965 | ret = logical_ring_wait_request(ringbuf, bytes); |
1253 | wa_ctx_emit(batch, index, (PIPE_CONTROL_FLUSH_L3 | |
- | 1254 | PIPE_CONTROL_GLOBAL_GTT_IVB | |
|
- | 1255 | PIPE_CONTROL_CS_STALL | |
|
- | 1256 | PIPE_CONTROL_QW_WRITE)); |
|
- | 1257 | wa_ctx_emit(batch, index, scratch_addr); |
|
- | 1258 | wa_ctx_emit(batch, index, 0); |
|
- | 1259 | wa_ctx_emit(batch, index, 0); |
|
- | 1260 | wa_ctx_emit(batch, index, 0); |
|
- | 1261 | ||
- | 1262 | /* Pad to end of cacheline */ |
|
- | 1263 | while (index % CACHELINE_DWORDS) |
|
- | 1264 | wa_ctx_emit(batch, index, MI_NOOP); |
|
- | 1265 | ||
- | 1266 | /* |
|
- | 1267 | * MI_BATCH_BUFFER_END is not required in Indirect ctx BB because |
|
- | 1268 | * execution depends on the length specified in terms of cache lines |
|
- | 1269 | * in the register CTX_RCS_INDIRECT_CTX |
|
- | 1270 | */ |
|
- | 1271 | ||
- | 1272 | return wa_ctx_end(wa_ctx, *offset = index, CACHELINE_DWORDS); |
|
- | 1273 | } |
|
966 | if (ret != -ENOSPC) |
1274 | |
- | 1275 | /** |
|
- | 1276 | * gen8_init_perctx_bb() - initialize per ctx batch with WA |
|
967 | return ret; |
1277 | * |
968 | 1278 | * @ring: only applicable for RCS |
|
969 | /* Force the context submission in case we have been skipping it */ |
- | |
970 | intel_logical_ring_advance_and_submit(ringbuf); |
- | |
971 | 1279 | * @wa_ctx: structure representing wa_ctx |
|
972 | /* With GEM the hangcheck timer should kick us out of the loop, |
- | |
973 | * leaving it early runs the risk of corrupting GEM state (due |
- | |
974 | * to running on almost untested codepaths). But on resume |
- | |
975 | * timers don't work yet, so prevent a complete hang in that |
- | |
976 | * case by choosing an insanely large timeout. */ |
1280 | * offset: specifies start of the batch, should be cache-aligned. |
977 | end = jiffies + 60 * HZ; |
1281 | * size: size of the batch in DWORDS but HW expects in terms of cachelines |
978 | 1282 | * @batch: page in which WA are loaded |
|
- | 1283 | * @offset: This field specifies the start of this batch. |
|
- | 1284 | * This batch is started immediately after indirect_ctx batch. Since we ensure |
|
- | 1285 | * that indirect_ctx ends on a cacheline this batch is aligned automatically. |
|
979 | do { |
1286 | * |
980 | ringbuf->head = I915_READ_HEAD(ring); |
1287 | * The number of DWORDS written are returned using this field. |
981 | ringbuf->space = intel_ring_space(ringbuf); |
1288 | * |
- | 1289 | * This batch is terminated with MI_BATCH_BUFFER_END and so we need not add padding |
|
982 | if (ringbuf->space >= bytes) { |
1290 | * to align it with cacheline as padding after MI_BATCH_BUFFER_END is redundant. |
- | 1291 | */ |
|
- | 1292 | static int gen8_init_perctx_bb(struct intel_engine_cs *ring, |
|
983 | ret = 0; |
1293 | struct i915_wa_ctx_bb *wa_ctx, |
- | 1294 | uint32_t *const batch, |
|
- | 1295 | uint32_t *offset) |
|
984 | break; |
1296 | { |
985 | } |
1297 | uint32_t index = wa_ctx_start(wa_ctx, *offset, CACHELINE_DWORDS); |
- | 1298 | ||
- | 1299 | /* WaDisableCtxRestoreArbitration:bdw,chv */ |
|
- | 1300 | wa_ctx_emit(batch, index, MI_ARB_ON_OFF | MI_ARB_ENABLE); |
|
- | 1301 | ||
- | 1302 | wa_ctx_emit(batch, index, MI_BATCH_BUFFER_END); |
|
- | 1303 | ||
- | 1304 | return wa_ctx_end(wa_ctx, *offset = index, 1); |
|
986 | 1305 | } |
|
987 | msleep(1); |
1306 | |
- | 1307 | static int gen9_init_indirectctx_bb(struct intel_engine_cs *ring, |
|
988 | 1308 | struct i915_wa_ctx_bb *wa_ctx, |
|
989 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, |
1309 | uint32_t *const batch, |
- | 1310 | uint32_t *offset) |
|
990 | dev_priv->mm.interruptible); |
1311 | { |
- | 1312 | int ret; |
|
- | 1313 | struct drm_device *dev = ring->dev; |
|
- | 1314 | uint32_t index = wa_ctx_start(wa_ctx, *offset, CACHELINE_DWORDS); |
|
- | 1315 | ||
- | 1316 | /* WaDisableCtxRestoreArbitration:skl,bxt */ |
|
- | 1317 | if ((IS_SKYLAKE(dev) && (INTEL_REVID(dev) <= SKL_REVID_D0)) || |
|
991 | if (ret) |
1318 | (IS_BROXTON(dev) && (INTEL_REVID(dev) == BXT_REVID_A0))) |
- | 1319 | wa_ctx_emit(batch, index, MI_ARB_ON_OFF | MI_ARB_DISABLE); |
|
992 | break; |
1320 | |
- | 1321 | /* WaFlushCoherentL3CacheLinesAtContextSwitch:skl,bxt */ |
|
- | 1322 | ret = gen8_emit_flush_coherentl3_wa(ring, batch, index); |
|
- | 1323 | if (ret < 0) |
|
- | 1324 | return ret; |
|
- | 1325 | index = ret; |
|
- | 1326 | ||
- | 1327 | /* Pad to end of cacheline */ |
|
- | 1328 | while (index % CACHELINE_DWORDS) |
|
993 | 1329 | wa_ctx_emit(batch, index, MI_NOOP); |
|
994 | if (time_after(jiffies, end)) { |
1330 | |
995 | ret = -EBUSY; |
1331 | return wa_ctx_end(wa_ctx, *offset = index, CACHELINE_DWORDS); |
996 | break; |
1332 | } |
997 | } |
1333 | |
998 | } while (1); |
1334 | static int gen9_init_perctx_bb(struct intel_engine_cs *ring, |
999 | 1335 | struct i915_wa_ctx_bb *wa_ctx, |
|
1000 | return ret; |
1336 | uint32_t *const batch, |
1001 | } |
1337 | uint32_t *offset) |
1002 | 1338 | { |
|
1003 | static int logical_ring_wrap_buffer(struct intel_ringbuffer *ringbuf) |
1339 | struct drm_device *dev = ring->dev; |
1004 | { |
1340 | uint32_t index = wa_ctx_start(wa_ctx, *offset, CACHELINE_DWORDS); |
1005 | uint32_t __iomem *virt; |
1341 | |
1006 | int rem = ringbuf->size - ringbuf->tail; |
1342 | /* WaSetDisablePixMaskCammingAndRhwoInCommonSliceChicken:skl,bxt */ |
1007 | 1343 | if ((IS_SKYLAKE(dev) && (INTEL_REVID(dev) <= SKL_REVID_B0)) || |
|
1008 | if (ringbuf->space < rem) { |
1344 | (IS_BROXTON(dev) && (INTEL_REVID(dev) == BXT_REVID_A0))) { |
1009 | int ret = logical_ring_wait_for_space(ringbuf, rem); |
1345 | wa_ctx_emit(batch, index, MI_LOAD_REGISTER_IMM(1)); |
1010 | 1346 | wa_ctx_emit(batch, index, GEN9_SLICE_COMMON_ECO_CHICKEN0); |
|
1011 | if (ret) |
1347 | wa_ctx_emit(batch, index, |
1012 | return ret; |
1348 | _MASKED_BIT_ENABLE(DISABLE_PIXEL_MASK_CAMMING)); |
1013 | } |
1349 | wa_ctx_emit(batch, index, MI_NOOP); |
1014 | 1350 | } |
|
1015 | virt = ringbuf->virtual_start + ringbuf->tail; |
1351 | |
1016 | rem /= 4; |
1352 | /* WaDisableCtxRestoreArbitration:skl,bxt */ |
1017 | while (rem--) |
1353 | if ((IS_SKYLAKE(dev) && (INTEL_REVID(dev) <= SKL_REVID_D0)) || |
1018 | iowrite32(MI_NOOP, virt++); |
1354 | (IS_BROXTON(dev) && (INTEL_REVID(dev) == BXT_REVID_A0))) |
1019 | 1355 | wa_ctx_emit(batch, index, MI_ARB_ON_OFF | MI_ARB_ENABLE); |
|
1020 | ringbuf->tail = 0; |
1356 | |
1021 | ringbuf->space = intel_ring_space(ringbuf); |
1357 | wa_ctx_emit(batch, index, MI_BATCH_BUFFER_END); |
1022 | 1358 | ||
1023 | return 0; |
1359 | return wa_ctx_end(wa_ctx, *offset = index, 1); |
1024 | } |
1360 | } |
1025 | 1361 | ||
1026 | static int logical_ring_prepare(struct intel_ringbuffer *ringbuf, int bytes) |
1362 | static int lrc_setup_wa_ctx_obj(struct intel_engine_cs *ring, u32 size) |
1027 | { |
1363 | { |
1028 | int ret; |
1364 | int ret; |
1029 | 1365 | ||
1030 | if (unlikely(ringbuf->tail + bytes > ringbuf->effective_size)) { |
1366 | ring->wa_ctx.obj = i915_gem_alloc_object(ring->dev, PAGE_ALIGN(size)); |
1031 | ret = logical_ring_wrap_buffer(ringbuf); |
1367 | if (!ring->wa_ctx.obj) { |
1032 | if (unlikely(ret)) |
1368 | DRM_DEBUG_DRIVER("alloc LRC WA ctx backing obj failed.\n"); |
1033 | return ret; |
1369 | return -ENOMEM; |
1034 | } |
1370 | } |
- | 1371 | ||
1035 | 1372 | ret = i915_gem_obj_ggtt_pin(ring->wa_ctx.obj, PAGE_SIZE, 0); |
|
1036 | if (unlikely(ringbuf->space < bytes)) { |
1373 | if (ret) { |
- | 1374 | DRM_DEBUG_DRIVER("pin LRC WA ctx backing obj failed: %d\n", |
|
1037 | ret = logical_ring_wait_for_space(ringbuf, bytes); |
1375 | ret); |
1038 | if (unlikely(ret)) |
1376 | drm_gem_object_unreference(&ring->wa_ctx.obj->base); |
1039 | return ret; |
1377 | return ret; |
1040 | } |
1378 | } |
1041 | 1379 | ||
1042 | return 0; |
1380 | return 0; |
1043 | } |
1381 | } |
1044 | - | ||
1045 | /** |
- | |
1046 | * intel_logical_ring_begin() - prepare the logical ringbuffer to accept some commands |
- | |
1047 | * |
- | |
1048 | * @ringbuf: Logical ringbuffer. |
- | |
1049 | * @num_dwords: number of DWORDs that we plan to write to the ringbuffer. |
- | |
1050 | * |
- | |
1051 | * The ringbuffer might not be ready to accept the commands right away (maybe it needs to |
- | |
1052 | * be wrapped, or wait a bit for the tail to be updated). This function takes care of that |
- | |
1053 | * and also preallocates a request (every workload submission is still mediated through |
- | |
1054 | * requests, same as it did with legacy ringbuffer submission). |
- | |
1055 | * |
- | |
1056 | * Return: non-zero if the ringbuffer is not ready to be written to. |
- | |
1057 | */ |
1382 | |
1058 | int intel_logical_ring_begin(struct intel_ringbuffer *ringbuf, int num_dwords) |
1383 | static void lrc_destroy_wa_ctx_obj(struct intel_engine_cs *ring) |
1059 | { |
1384 | { |
1060 | struct intel_engine_cs *ring = ringbuf->ring; |
1385 | if (ring->wa_ctx.obj) { |
1061 | struct drm_device *dev = ring->dev; |
1386 | i915_gem_object_ggtt_unpin(ring->wa_ctx.obj); |
1062 | struct drm_i915_private *dev_priv = dev->dev_private; |
1387 | drm_gem_object_unreference(&ring->wa_ctx.obj->base); |
1063 | int ret; |
1388 | ring->wa_ctx.obj = NULL; |
1064 | - | ||
1065 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, |
- | |
1066 | dev_priv->mm.interruptible); |
1389 | } |
1067 | if (ret) |
- | |
1068 | return ret; |
1390 | } |
- | 1391 | ||
1069 | 1392 | static int intel_init_workaround_bb(struct intel_engine_cs *ring) |
|
- | 1393 | { |
|
1070 | ret = logical_ring_prepare(ringbuf, num_dwords * sizeof(uint32_t)); |
1394 | int ret; |
- | 1395 | uint32_t *batch; |
|
- | 1396 | uint32_t offset; |
|
1071 | if (ret) |
- | |
1072 | return ret; |
- | |
1073 | - | ||
1074 | /* Preallocate the olr before touching the ring */ |
1397 | struct page *page; |
1075 | ret = logical_ring_alloc_seqno(ring, ringbuf->FIXME_lrc_ctx); |
1398 | struct i915_ctx_workarounds *wa_ctx = &ring->wa_ctx; |
- | 1399 | ||
- | 1400 | WARN_ON(ring->id != RCS); |
|
- | 1401 | ||
1076 | if (ret) |
1402 | /* update this when WA for higher Gen are added */ |
1077 | return ret; |
1403 | if (INTEL_INFO(ring->dev)->gen > 9) { |
1078 | 1404 | DRM_ERROR("WA batch buffer is not initialized for Gen%d\n", |
|
1079 | ringbuf->space -= num_dwords * sizeof(uint32_t); |
1405 | INTEL_INFO(ring->dev)->gen); |
1080 | return 0; |
1406 | return 0; |
1081 | } |
1407 | } |
1082 | 1408 | ||
1083 | static int intel_logical_ring_workarounds_emit(struct intel_engine_cs *ring, |
1409 | /* some WA perform writes to scratch page, ensure it is valid */ |
1084 | struct intel_context *ctx) |
- | |
1085 | { |
- | |
1086 | int ret, i; |
1410 | if (ring->scratch.obj == NULL) { |
1087 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
1411 | DRM_ERROR("scratch page not allocated for %s\n", ring->name); |
1088 | struct drm_device *dev = ring->dev; |
- | |
1089 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
- | 1412 | return -EINVAL; |
|
1090 | struct i915_workarounds *w = &dev_priv->workarounds; |
1413 | } |
- | 1414 | ||
- | 1415 | ret = lrc_setup_wa_ctx_obj(ring, PAGE_SIZE); |
|
1091 | 1416 | if (ret) { |
|
- | 1417 | DRM_DEBUG_DRIVER("Failed to setup context WA page: %d\n", ret); |
|
- | 1418 | return ret; |
|
1092 | if (WARN_ON(w->count == 0)) |
1419 | } |
- | 1420 | ||
- | 1421 | page = i915_gem_object_get_page(wa_ctx->obj, 0); |
|
- | 1422 | batch = kmap_atomic(page); |
|
1093 | return 0; |
1423 | offset = 0; |
- | 1424 | ||
- | 1425 | if (INTEL_INFO(ring->dev)->gen == 8) { |
|
- | 1426 | ret = gen8_init_indirectctx_bb(ring, |
|
1094 | 1427 | &wa_ctx->indirect_ctx, |
|
1095 | ring->gpu_caches_dirty = true; |
1428 | batch, |
1096 | ret = logical_ring_flush_all_caches(ringbuf); |
1429 | &offset); |
- | 1430 | if (ret) |
|
- | 1431 | goto out; |
|
- | 1432 | ||
1097 | if (ret) |
1433 | ret = gen8_init_perctx_bb(ring, |
1098 | return ret; |
1434 | &wa_ctx->per_ctx, |
- | 1435 | batch, |
|
- | 1436 | &offset); |
|
- | 1437 | if (ret) |
|
- | 1438 | goto out; |
|
- | 1439 | } else if (INTEL_INFO(ring->dev)->gen == 9) { |
|
- | 1440 | ret = gen9_init_indirectctx_bb(ring, |
|
- | 1441 | &wa_ctx->indirect_ctx, |
|
1099 | 1442 | batch, |
|
1100 | ret = intel_logical_ring_begin(ringbuf, w->count * 2 + 2); |
1443 | &offset); |
- | 1444 | if (ret) |
|
1101 | if (ret) |
1445 | goto out; |
- | 1446 | ||
1102 | return ret; |
1447 | ret = gen9_init_perctx_bb(ring, |
1103 | 1448 | &wa_ctx->per_ctx, |
|
1104 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(w->count)); |
- | |
1105 | for (i = 0; i < w->count; i++) { |
- | |
1106 | intel_logical_ring_emit(ringbuf, w->reg[i].addr); |
1449 | batch, |
1107 | intel_logical_ring_emit(ringbuf, w->reg[i].value); |
1450 | &offset); |
1108 | } |
- | |
1109 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
1451 | if (ret) |
1110 | 1452 | goto out; |
|
1111 | intel_logical_ring_advance(ringbuf); |
1453 | } |
1112 | 1454 | ||
1113 | ring->gpu_caches_dirty = true; |
1455 | out: |
1114 | ret = logical_ring_flush_all_caches(ringbuf); |
1456 | kunmap_atomic(batch); |
1115 | if (ret) |
1457 | if (ret) |
1116 | return ret; |
1458 | lrc_destroy_wa_ctx_obj(ring); |
1117 | 1459 | ||
1118 | return 0; |
1460 | return ret; |
1119 | } |
1461 | } |
1120 | 1462 | ||
1121 | static int gen8_init_common_ring(struct intel_engine_cs *ring) |
1463 | static int gen8_init_common_ring(struct intel_engine_cs *ring) |
1122 | { |
1464 | { |
1123 | struct drm_device *dev = ring->dev; |
1465 | struct drm_device *dev = ring->dev; |
1124 | struct drm_i915_private *dev_priv = dev->dev_private; |
1466 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | 1467 | u8 next_context_status_buffer_hw; |
|
- | 1468 | ||
- | 1469 | lrc_setup_hardware_status_page(ring, |
|
- | 1470 | ring->default_context->engine[ring->id].state); |
|
1125 | 1471 | ||
1126 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
1472 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
1127 | I915_WRITE(RING_HWSTAM(ring->mmio_base), 0xffffffff); |
1473 | I915_WRITE(RING_HWSTAM(ring->mmio_base), 0xffffffff); |
- | 1474 | ||
- | 1475 | if (ring->status_page.obj) { |
|
- | 1476 | I915_WRITE(RING_HWS_PGA(ring->mmio_base), |
|
- | 1477 | (u32)ring->status_page.gfx_addr); |
|
- | 1478 | POSTING_READ(RING_HWS_PGA(ring->mmio_base)); |
|
- | 1479 | } |
|
1128 | 1480 | ||
1129 | I915_WRITE(RING_MODE_GEN7(ring), |
1481 | I915_WRITE(RING_MODE_GEN7(ring), |
1130 | _MASKED_BIT_DISABLE(GFX_REPLAY_MODE) | |
1482 | _MASKED_BIT_DISABLE(GFX_REPLAY_MODE) | |
1131 | _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); |
1483 | _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); |
1132 | POSTING_READ(RING_MODE_GEN7(ring)); |
1484 | POSTING_READ(RING_MODE_GEN7(ring)); |
- | 1485 | ||
- | 1486 | /* |
|
- | 1487 | * Instead of resetting the Context Status Buffer (CSB) read pointer to |
|
- | 1488 | * zero, we need to read the write pointer from hardware and use its |
|
- | 1489 | * value because "this register is power context save restored". |
|
- | 1490 | * Effectively, these states have been observed: |
|
- | 1491 | * |
|
- | 1492 | * | Suspend-to-idle (freeze) | Suspend-to-RAM (mem) | |
|
- | 1493 | * BDW | CSB regs not reset | CSB regs reset | |
|
- | 1494 | * CHT | CSB regs not reset | CSB regs not reset | |
|
- | 1495 | */ |
|
- | 1496 | next_context_status_buffer_hw = (I915_READ(RING_CONTEXT_STATUS_PTR(ring)) |
|
- | 1497 | & GEN8_CSB_PTR_MASK); |
|
- | 1498 | ||
- | 1499 | /* |
|
- | 1500 | * When the CSB registers are reset (also after power-up / gpu reset), |
|
- | 1501 | * CSB write pointer is set to all 1's, which is not valid, use '5' in |
|
- | 1502 | * this special case, so the first element read is CSB[0]. |
|
- | 1503 | */ |
|
- | 1504 | if (next_context_status_buffer_hw == GEN8_CSB_PTR_MASK) |
|
- | 1505 | next_context_status_buffer_hw = (GEN8_CSB_ENTRIES - 1); |
|
- | 1506 | ||
- | 1507 | ring->next_context_status_buffer = next_context_status_buffer_hw; |
|
1133 | DRM_DEBUG_DRIVER("Execlists enabled for %s\n", ring->name); |
1508 | DRM_DEBUG_DRIVER("Execlists enabled for %s\n", ring->name); |
1134 | 1509 | ||
1135 | memset(&ring->hangcheck, 0, sizeof(ring->hangcheck)); |
1510 | memset(&ring->hangcheck, 0, sizeof(ring->hangcheck)); |
1136 | 1511 | ||
1137 | return 0; |
1512 | return 0; |
1138 | } |
1513 | } |
1139 | 1514 | ||
1140 | static int gen8_init_render_ring(struct intel_engine_cs *ring) |
1515 | static int gen8_init_render_ring(struct intel_engine_cs *ring) |
1141 | { |
1516 | { |
1142 | struct drm_device *dev = ring->dev; |
1517 | struct drm_device *dev = ring->dev; |
1143 | struct drm_i915_private *dev_priv = dev->dev_private; |
1518 | struct drm_i915_private *dev_priv = dev->dev_private; |
1144 | int ret; |
1519 | int ret; |
1145 | 1520 | ||
1146 | ret = gen8_init_common_ring(ring); |
1521 | ret = gen8_init_common_ring(ring); |
1147 | if (ret) |
1522 | if (ret) |
1148 | return ret; |
1523 | return ret; |
1149 | 1524 | ||
1150 | /* We need to disable the AsyncFlip performance optimisations in order |
1525 | /* We need to disable the AsyncFlip performance optimisations in order |
1151 | * to use MI_WAIT_FOR_EVENT within the CS. It should already be |
1526 | * to use MI_WAIT_FOR_EVENT within the CS. It should already be |
1152 | * programmed to '1' on all products. |
1527 | * programmed to '1' on all products. |
1153 | * |
1528 | * |
1154 | * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv,bdw,chv |
1529 | * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv,bdw,chv |
1155 | */ |
1530 | */ |
1156 | I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(ASYNC_FLIP_PERF_DISABLE)); |
1531 | I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(ASYNC_FLIP_PERF_DISABLE)); |
- | 1532 | ||
- | 1533 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); |
|
- | 1534 | ||
- | 1535 | return init_workarounds_ring(ring); |
|
- | 1536 | } |
|
- | 1537 | ||
- | 1538 | static int gen9_init_render_ring(struct intel_engine_cs *ring) |
|
- | 1539 | { |
|
- | 1540 | int ret; |
|
1157 | 1541 | ||
1158 | ret = intel_init_pipe_control(ring); |
1542 | ret = gen8_init_common_ring(ring); |
1159 | if (ret) |
1543 | if (ret) |
1160 | return ret; |
1544 | return ret; |
1161 | - | ||
1162 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); |
- | |
1163 | 1545 | ||
1164 | return init_workarounds_ring(ring); |
1546 | return init_workarounds_ring(ring); |
1165 | } |
1547 | } |
1166 | 1548 | ||
1167 | static int gen8_emit_bb_start(struct intel_ringbuffer *ringbuf, |
- | |
1168 | u64 offset, unsigned flags) |
1549 | static int intel_logical_ring_emit_pdps(struct drm_i915_gem_request *req) |
- | 1550 | { |
|
- | 1551 | struct i915_hw_ppgtt *ppgtt = req->ctx->ppgtt; |
|
- | 1552 | struct intel_engine_cs *ring = req->ring; |
|
- | 1553 | struct intel_ringbuffer *ringbuf = req->ringbuf; |
|
- | 1554 | const int num_lri_cmds = GEN8_LEGACY_PDPES * 2; |
|
- | 1555 | int i, ret; |
|
- | 1556 | ||
- | 1557 | ret = intel_logical_ring_begin(req, num_lri_cmds * 2 + 2); |
|
- | 1558 | if (ret) |
|
- | 1559 | return ret; |
|
- | 1560 | ||
- | 1561 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(num_lri_cmds)); |
|
- | 1562 | for (i = GEN8_LEGACY_PDPES - 1; i >= 0; i--) { |
|
- | 1563 | const dma_addr_t pd_daddr = i915_page_dir_dma_addr(ppgtt, i); |
|
- | 1564 | ||
- | 1565 | intel_logical_ring_emit(ringbuf, GEN8_RING_PDP_UDW(ring, i)); |
|
- | 1566 | intel_logical_ring_emit(ringbuf, upper_32_bits(pd_daddr)); |
|
- | 1567 | intel_logical_ring_emit(ringbuf, GEN8_RING_PDP_LDW(ring, i)); |
|
- | 1568 | intel_logical_ring_emit(ringbuf, lower_32_bits(pd_daddr)); |
|
- | 1569 | } |
|
- | 1570 | ||
- | 1571 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
|
- | 1572 | intel_logical_ring_advance(ringbuf); |
|
- | 1573 | ||
- | 1574 | return 0; |
|
- | 1575 | } |
|
- | 1576 | ||
- | 1577 | static int gen8_emit_bb_start(struct drm_i915_gem_request *req, |
|
- | 1578 | u64 offset, unsigned dispatch_flags) |
|
- | 1579 | { |
|
1169 | { |
1580 | struct intel_ringbuffer *ringbuf = req->ringbuf; |
1170 | bool ppgtt = !(flags & I915_DISPATCH_SECURE); |
1581 | bool ppgtt = !(dispatch_flags & I915_DISPATCH_SECURE); |
- | 1582 | int ret; |
|
- | 1583 | ||
- | 1584 | /* Don't rely in hw updating PDPs, specially in lite-restore. |
|
- | 1585 | * Ideally, we should set Force PD Restore in ctx descriptor, |
|
- | 1586 | * but we can't. Force Restore would be a second option, but |
|
- | 1587 | * it is unsafe in case of lite-restore (because the ctx is |
|
- | 1588 | * not idle). PML4 is allocated during ppgtt init so this is |
|
- | 1589 | * not needed in 48-bit.*/ |
|
- | 1590 | if (req->ctx->ppgtt && |
|
- | 1591 | (intel_ring_flag(req->ring) & req->ctx->ppgtt->pd_dirty_rings)) { |
|
- | 1592 | if (!USES_FULL_48BIT_PPGTT(req->i915) && |
|
- | 1593 | !intel_vgpu_active(req->i915->dev)) { |
|
- | 1594 | ret = intel_logical_ring_emit_pdps(req); |
|
- | 1595 | if (ret) |
|
- | 1596 | return ret; |
|
- | 1597 | } |
|
- | 1598 | ||
- | 1599 | req->ctx->ppgtt->pd_dirty_rings &= ~intel_ring_flag(req->ring); |
|
1171 | int ret; |
1600 | } |
1172 | 1601 | ||
1173 | ret = intel_logical_ring_begin(ringbuf, 4); |
1602 | ret = intel_logical_ring_begin(req, 4); |
1174 | if (ret) |
1603 | if (ret) |
1175 | return ret; |
1604 | return ret; |
1176 | 1605 | ||
1177 | /* FIXME(BDW): Address space and security selectors. */ |
1606 | /* FIXME(BDW): Address space and security selectors. */ |
1178 | intel_logical_ring_emit(ringbuf, MI_BATCH_BUFFER_START_GEN8 | (ppgtt<<8)); |
1607 | intel_logical_ring_emit(ringbuf, MI_BATCH_BUFFER_START_GEN8 | |
- | 1608 | (ppgtt<<8) | |
|
- | 1609 | (dispatch_flags & I915_DISPATCH_RS ? |
|
- | 1610 | MI_BATCH_RESOURCE_STREAMER : 0)); |
|
1179 | intel_logical_ring_emit(ringbuf, lower_32_bits(offset)); |
1611 | intel_logical_ring_emit(ringbuf, lower_32_bits(offset)); |
1180 | intel_logical_ring_emit(ringbuf, upper_32_bits(offset)); |
1612 | intel_logical_ring_emit(ringbuf, upper_32_bits(offset)); |
1181 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
1613 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
1182 | intel_logical_ring_advance(ringbuf); |
1614 | intel_logical_ring_advance(ringbuf); |
1183 | 1615 | ||
1184 | return 0; |
1616 | return 0; |
1185 | } |
1617 | } |
1186 | 1618 | ||
1187 | static bool gen8_logical_ring_get_irq(struct intel_engine_cs *ring) |
1619 | static bool gen8_logical_ring_get_irq(struct intel_engine_cs *ring) |
1188 | { |
1620 | { |
1189 | struct drm_device *dev = ring->dev; |
1621 | struct drm_device *dev = ring->dev; |
1190 | struct drm_i915_private *dev_priv = dev->dev_private; |
1622 | struct drm_i915_private *dev_priv = dev->dev_private; |
1191 | unsigned long flags; |
1623 | unsigned long flags; |
1192 | 1624 | ||
1193 | if (WARN_ON(!intel_irqs_enabled(dev_priv))) |
1625 | if (WARN_ON(!intel_irqs_enabled(dev_priv))) |
1194 | return false; |
1626 | return false; |
1195 | 1627 | ||
1196 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
1628 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
1197 | if (ring->irq_refcount++ == 0) { |
1629 | if (ring->irq_refcount++ == 0) { |
1198 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
1630 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
1199 | POSTING_READ(RING_IMR(ring->mmio_base)); |
1631 | POSTING_READ(RING_IMR(ring->mmio_base)); |
1200 | } |
1632 | } |
1201 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
1633 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
1202 | 1634 | ||
1203 | return true; |
1635 | return true; |
1204 | } |
1636 | } |
1205 | 1637 | ||
1206 | static void gen8_logical_ring_put_irq(struct intel_engine_cs *ring) |
1638 | static void gen8_logical_ring_put_irq(struct intel_engine_cs *ring) |
1207 | { |
1639 | { |
1208 | struct drm_device *dev = ring->dev; |
1640 | struct drm_device *dev = ring->dev; |
1209 | struct drm_i915_private *dev_priv = dev->dev_private; |
1641 | struct drm_i915_private *dev_priv = dev->dev_private; |
1210 | unsigned long flags; |
1642 | unsigned long flags; |
1211 | 1643 | ||
1212 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
1644 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
1213 | if (--ring->irq_refcount == 0) { |
1645 | if (--ring->irq_refcount == 0) { |
1214 | I915_WRITE_IMR(ring, ~ring->irq_keep_mask); |
1646 | I915_WRITE_IMR(ring, ~ring->irq_keep_mask); |
1215 | POSTING_READ(RING_IMR(ring->mmio_base)); |
1647 | POSTING_READ(RING_IMR(ring->mmio_base)); |
1216 | } |
1648 | } |
1217 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
1649 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
1218 | } |
1650 | } |
1219 | 1651 | ||
1220 | static int gen8_emit_flush(struct intel_ringbuffer *ringbuf, |
1652 | static int gen8_emit_flush(struct drm_i915_gem_request *request, |
1221 | u32 invalidate_domains, |
1653 | u32 invalidate_domains, |
1222 | u32 unused) |
1654 | u32 unused) |
1223 | { |
1655 | { |
- | 1656 | struct intel_ringbuffer *ringbuf = request->ringbuf; |
|
1224 | struct intel_engine_cs *ring = ringbuf->ring; |
1657 | struct intel_engine_cs *ring = ringbuf->ring; |
1225 | struct drm_device *dev = ring->dev; |
1658 | struct drm_device *dev = ring->dev; |
1226 | struct drm_i915_private *dev_priv = dev->dev_private; |
1659 | struct drm_i915_private *dev_priv = dev->dev_private; |
1227 | uint32_t cmd; |
1660 | uint32_t cmd; |
1228 | int ret; |
1661 | int ret; |
1229 | 1662 | ||
1230 | ret = intel_logical_ring_begin(ringbuf, 4); |
1663 | ret = intel_logical_ring_begin(request, 4); |
1231 | if (ret) |
1664 | if (ret) |
1232 | return ret; |
1665 | return ret; |
1233 | 1666 | ||
1234 | cmd = MI_FLUSH_DW + 1; |
1667 | cmd = MI_FLUSH_DW + 1; |
1235 | 1668 | ||
1236 | if (ring == &dev_priv->ring[VCS]) { |
1669 | /* We always require a command barrier so that subsequent |
1237 | if (invalidate_domains & I915_GEM_GPU_DOMAINS) |
1670 | * commands, such as breadcrumb interrupts, are strictly ordered |
1238 | cmd |= MI_INVALIDATE_TLB | MI_INVALIDATE_BSD | |
1671 | * wrt the contents of the write cache being flushed to memory |
- | 1672 | * (and thus being coherent from the CPU). |
|
1239 | MI_FLUSH_DW_STORE_INDEX | |
1673 | */ |
1240 | MI_FLUSH_DW_OP_STOREDW; |
1674 | cmd |= MI_FLUSH_DW_STORE_INDEX | MI_FLUSH_DW_OP_STOREDW; |
1241 | } else { |
1675 | |
1242 | if (invalidate_domains & I915_GEM_DOMAIN_RENDER) |
1676 | if (invalidate_domains & I915_GEM_GPU_DOMAINS) { |
- | 1677 | cmd |= MI_INVALIDATE_TLB; |
|
1243 | cmd |= MI_INVALIDATE_TLB | MI_FLUSH_DW_STORE_INDEX | |
1678 | if (ring == &dev_priv->ring[VCS]) |
1244 | MI_FLUSH_DW_OP_STOREDW; |
1679 | cmd |= MI_INVALIDATE_BSD; |
1245 | } |
1680 | } |
1246 | 1681 | ||
1247 | intel_logical_ring_emit(ringbuf, cmd); |
1682 | intel_logical_ring_emit(ringbuf, cmd); |
1248 | intel_logical_ring_emit(ringbuf, |
1683 | intel_logical_ring_emit(ringbuf, |
1249 | I915_GEM_HWS_SCRATCH_ADDR | |
1684 | I915_GEM_HWS_SCRATCH_ADDR | |
1250 | MI_FLUSH_DW_USE_GTT); |
1685 | MI_FLUSH_DW_USE_GTT); |
1251 | intel_logical_ring_emit(ringbuf, 0); /* upper addr */ |
1686 | intel_logical_ring_emit(ringbuf, 0); /* upper addr */ |
1252 | intel_logical_ring_emit(ringbuf, 0); /* value */ |
1687 | intel_logical_ring_emit(ringbuf, 0); /* value */ |
1253 | intel_logical_ring_advance(ringbuf); |
1688 | intel_logical_ring_advance(ringbuf); |
1254 | 1689 | ||
1255 | return 0; |
1690 | return 0; |
1256 | } |
1691 | } |
1257 | 1692 | ||
1258 | static int gen8_emit_flush_render(struct intel_ringbuffer *ringbuf, |
1693 | static int gen8_emit_flush_render(struct drm_i915_gem_request *request, |
1259 | u32 invalidate_domains, |
1694 | u32 invalidate_domains, |
1260 | u32 flush_domains) |
1695 | u32 flush_domains) |
1261 | { |
1696 | { |
- | 1697 | struct intel_ringbuffer *ringbuf = request->ringbuf; |
|
1262 | struct intel_engine_cs *ring = ringbuf->ring; |
1698 | struct intel_engine_cs *ring = ringbuf->ring; |
1263 | u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES; |
1699 | u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES; |
- | 1700 | bool vf_flush_wa; |
|
1264 | u32 flags = 0; |
1701 | u32 flags = 0; |
1265 | int ret; |
1702 | int ret; |
1266 | 1703 | ||
1267 | flags |= PIPE_CONTROL_CS_STALL; |
1704 | flags |= PIPE_CONTROL_CS_STALL; |
1268 | 1705 | ||
1269 | if (flush_domains) { |
1706 | if (flush_domains) { |
1270 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; |
1707 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; |
1271 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; |
1708 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; |
- | 1709 | flags |= PIPE_CONTROL_FLUSH_ENABLE; |
|
1272 | } |
1710 | } |
1273 | 1711 | ||
1274 | if (invalidate_domains) { |
1712 | if (invalidate_domains) { |
1275 | flags |= PIPE_CONTROL_TLB_INVALIDATE; |
1713 | flags |= PIPE_CONTROL_TLB_INVALIDATE; |
1276 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; |
1714 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; |
1277 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; |
1715 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; |
1278 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; |
1716 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; |
1279 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; |
1717 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; |
1280 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; |
1718 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; |
1281 | flags |= PIPE_CONTROL_QW_WRITE; |
1719 | flags |= PIPE_CONTROL_QW_WRITE; |
1282 | flags |= PIPE_CONTROL_GLOBAL_GTT_IVB; |
1720 | flags |= PIPE_CONTROL_GLOBAL_GTT_IVB; |
1283 | } |
1721 | } |
- | 1722 | ||
- | 1723 | /* |
|
- | 1724 | * On GEN9+ Before VF_CACHE_INVALIDATE we need to emit a NULL pipe |
|
- | 1725 | * control. |
|
- | 1726 | */ |
|
- | 1727 | vf_flush_wa = INTEL_INFO(ring->dev)->gen >= 9 && |
|
- | 1728 | flags & PIPE_CONTROL_VF_CACHE_INVALIDATE; |
|
1284 | 1729 | ||
1285 | ret = intel_logical_ring_begin(ringbuf, 6); |
1730 | ret = intel_logical_ring_begin(request, vf_flush_wa ? 12 : 6); |
1286 | if (ret) |
1731 | if (ret) |
1287 | return ret; |
1732 | return ret; |
- | 1733 | ||
- | 1734 | if (vf_flush_wa) { |
|
- | 1735 | intel_logical_ring_emit(ringbuf, GFX_OP_PIPE_CONTROL(6)); |
|
- | 1736 | intel_logical_ring_emit(ringbuf, 0); |
|
- | 1737 | intel_logical_ring_emit(ringbuf, 0); |
|
- | 1738 | intel_logical_ring_emit(ringbuf, 0); |
|
- | 1739 | intel_logical_ring_emit(ringbuf, 0); |
|
- | 1740 | intel_logical_ring_emit(ringbuf, 0); |
|
- | 1741 | } |
|
1288 | 1742 | ||
1289 | intel_logical_ring_emit(ringbuf, GFX_OP_PIPE_CONTROL(6)); |
1743 | intel_logical_ring_emit(ringbuf, GFX_OP_PIPE_CONTROL(6)); |
1290 | intel_logical_ring_emit(ringbuf, flags); |
1744 | intel_logical_ring_emit(ringbuf, flags); |
1291 | intel_logical_ring_emit(ringbuf, scratch_addr); |
1745 | intel_logical_ring_emit(ringbuf, scratch_addr); |
1292 | intel_logical_ring_emit(ringbuf, 0); |
1746 | intel_logical_ring_emit(ringbuf, 0); |
1293 | intel_logical_ring_emit(ringbuf, 0); |
1747 | intel_logical_ring_emit(ringbuf, 0); |
1294 | intel_logical_ring_emit(ringbuf, 0); |
1748 | intel_logical_ring_emit(ringbuf, 0); |
1295 | intel_logical_ring_advance(ringbuf); |
1749 | intel_logical_ring_advance(ringbuf); |
1296 | 1750 | ||
1297 | return 0; |
1751 | return 0; |
1298 | } |
1752 | } |
1299 | 1753 | ||
1300 | static u32 gen8_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency) |
1754 | static u32 gen8_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency) |
1301 | { |
1755 | { |
1302 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); |
1756 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); |
1303 | } |
1757 | } |
1304 | 1758 | ||
1305 | static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno) |
1759 | static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno) |
1306 | { |
1760 | { |
1307 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); |
1761 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); |
1308 | } |
1762 | } |
- | 1763 | ||
- | 1764 | static u32 bxt_a_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency) |
|
- | 1765 | { |
|
- | 1766 | ||
- | 1767 | /* |
|
- | 1768 | * On BXT A steppings there is a HW coherency issue whereby the |
|
- | 1769 | * MI_STORE_DATA_IMM storing the completed request's seqno |
|
- | 1770 | * occasionally doesn't invalidate the CPU cache. Work around this by |
|
- | 1771 | * clflushing the corresponding cacheline whenever the caller wants |
|
- | 1772 | * the coherency to be guaranteed. Note that this cacheline is known |
|
- | 1773 | * to be clean at this point, since we only write it in |
|
- | 1774 | * bxt_a_set_seqno(), where we also do a clflush after the write. So |
|
- | 1775 | * this clflush in practice becomes an invalidate operation. |
|
- | 1776 | */ |
|
- | 1777 | ||
- | 1778 | if (!lazy_coherency) |
|
- | 1779 | intel_flush_status_page(ring, I915_GEM_HWS_INDEX); |
|
- | 1780 | ||
- | 1781 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); |
|
- | 1782 | } |
|
- | 1783 | ||
- | 1784 | static void bxt_a_set_seqno(struct intel_engine_cs *ring, u32 seqno) |
|
- | 1785 | { |
|
- | 1786 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); |
|
- | 1787 | ||
- | 1788 | /* See bxt_a_get_seqno() explaining the reason for the clflush. */ |
|
- | 1789 | intel_flush_status_page(ring, I915_GEM_HWS_INDEX); |
|
- | 1790 | } |
|
1309 | 1791 | ||
1310 | static int gen8_emit_request(struct intel_ringbuffer *ringbuf) |
1792 | static int gen8_emit_request(struct drm_i915_gem_request *request) |
- | 1793 | { |
|
1311 | { |
1794 | struct intel_ringbuffer *ringbuf = request->ringbuf; |
1312 | struct intel_engine_cs *ring = ringbuf->ring; |
1795 | struct intel_engine_cs *ring = ringbuf->ring; |
1313 | u32 cmd; |
1796 | u32 cmd; |
1314 | int ret; |
1797 | int ret; |
- | 1798 | ||
- | 1799 | /* |
|
- | 1800 | * Reserve space for 2 NOOPs at the end of each request to be |
|
- | 1801 | * used as a workaround for not being allowed to do lite |
|
- | 1802 | * restore with HEAD==TAIL (WaIdleLiteRestore). |
|
1315 | 1803 | */ |
|
1316 | ret = intel_logical_ring_begin(ringbuf, 6); |
1804 | ret = intel_logical_ring_begin(request, 8); |
1317 | if (ret) |
1805 | if (ret) |
1318 | return ret; |
1806 | return ret; |
1319 | 1807 | ||
1320 | cmd = MI_STORE_DWORD_IMM_GEN8; |
1808 | cmd = MI_STORE_DWORD_IMM_GEN4; |
1321 | cmd |= MI_GLOBAL_GTT; |
1809 | cmd |= MI_GLOBAL_GTT; |
1322 | 1810 | ||
1323 | intel_logical_ring_emit(ringbuf, cmd); |
1811 | intel_logical_ring_emit(ringbuf, cmd); |
1324 | intel_logical_ring_emit(ringbuf, |
1812 | intel_logical_ring_emit(ringbuf, |
1325 | (ring->status_page.gfx_addr + |
1813 | (ring->status_page.gfx_addr + |
1326 | (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT))); |
1814 | (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT))); |
1327 | intel_logical_ring_emit(ringbuf, 0); |
1815 | intel_logical_ring_emit(ringbuf, 0); |
1328 | intel_logical_ring_emit(ringbuf, ring->outstanding_lazy_seqno); |
1816 | intel_logical_ring_emit(ringbuf, i915_gem_request_get_seqno(request)); |
1329 | intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT); |
1817 | intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT); |
1330 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
1818 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
1331 | intel_logical_ring_advance_and_submit(ringbuf); |
1819 | intel_logical_ring_advance_and_submit(request); |
- | 1820 | ||
- | 1821 | /* |
|
- | 1822 | * Here we add two extra NOOPs as padding to avoid |
|
- | 1823 | * lite restore of a context with HEAD==TAIL. |
|
- | 1824 | */ |
|
- | 1825 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
|
- | 1826 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
|
- | 1827 | intel_logical_ring_advance(ringbuf); |
|
- | 1828 | ||
- | 1829 | return 0; |
|
- | 1830 | } |
|
- | 1831 | ||
- | 1832 | static int intel_lr_context_render_state_init(struct drm_i915_gem_request *req) |
|
- | 1833 | { |
|
- | 1834 | struct render_state so; |
|
- | 1835 | int ret; |
|
- | 1836 | ||
- | 1837 | ret = i915_gem_render_state_prepare(req->ring, &so); |
|
- | 1838 | if (ret) |
|
- | 1839 | return ret; |
|
- | 1840 | ||
1332 | 1841 | if (so.rodata == NULL) |
|
- | 1842 | return 0; |
|
- | 1843 | ||
- | 1844 | ret = req->ring->emit_bb_start(req, so.ggtt_offset, |
|
- | 1845 | I915_DISPATCH_SECURE); |
|
- | 1846 | if (ret) |
|
- | 1847 | goto out; |
|
- | 1848 | ||
- | 1849 | ret = req->ring->emit_bb_start(req, |
|
- | 1850 | (so.ggtt_offset + so.aux_batch_offset), |
|
- | 1851 | I915_DISPATCH_SECURE); |
|
- | 1852 | if (ret) |
|
- | 1853 | goto out; |
|
- | 1854 | ||
- | 1855 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req); |
|
- | 1856 | ||
- | 1857 | out: |
|
- | 1858 | i915_gem_render_state_fini(&so); |
|
- | 1859 | return ret; |
|
- | 1860 | } |
|
- | 1861 | ||
- | 1862 | static int gen8_init_rcs_context(struct drm_i915_gem_request *req) |
|
- | 1863 | { |
|
- | 1864 | int ret; |
|
- | 1865 | ||
- | 1866 | ret = intel_logical_ring_workarounds_emit(req); |
|
- | 1867 | if (ret) |
|
- | 1868 | return ret; |
|
- | 1869 | ||
- | 1870 | ret = intel_rcs_context_init_mocs(req); |
|
- | 1871 | /* |
|
- | 1872 | * Failing to program the MOCS is non-fatal.The system will not |
|
- | 1873 | * run at peak performance. So generate an error and carry on. |
|
- | 1874 | */ |
|
- | 1875 | if (ret) |
|
- | 1876 | DRM_ERROR("MOCS failed to program: expect performance issues.\n"); |
|
- | 1877 | ||
1333 | return 0; |
1878 | return intel_lr_context_render_state_init(req); |
1334 | } |
1879 | } |
1335 | 1880 | ||
1336 | /** |
1881 | /** |
1337 | * intel_logical_ring_cleanup() - deallocate the Engine Command Streamer |
1882 | * intel_logical_ring_cleanup() - deallocate the Engine Command Streamer |
1338 | * |
1883 | * |
1339 | * @ring: Engine Command Streamer. |
1884 | * @ring: Engine Command Streamer. |
1340 | * |
1885 | * |
1341 | */ |
1886 | */ |
1342 | void intel_logical_ring_cleanup(struct intel_engine_cs *ring) |
1887 | void intel_logical_ring_cleanup(struct intel_engine_cs *ring) |
1343 | { |
1888 | { |
1344 | struct drm_i915_private *dev_priv; |
1889 | struct drm_i915_private *dev_priv; |
1345 | 1890 | ||
1346 | if (!intel_ring_initialized(ring)) |
1891 | if (!intel_ring_initialized(ring)) |
1347 | return; |
1892 | return; |
1348 | 1893 | ||
1349 | dev_priv = ring->dev->dev_private; |
1894 | dev_priv = ring->dev->dev_private; |
1350 | 1895 | ||
1351 | intel_logical_ring_stop(ring); |
1896 | intel_logical_ring_stop(ring); |
1352 | WARN_ON((I915_READ_MODE(ring) & MODE_IDLE) == 0); |
1897 | WARN_ON((I915_READ_MODE(ring) & MODE_IDLE) == 0); |
1353 | ring->preallocated_lazy_request = NULL; |
- | |
1354 | ring->outstanding_lazy_seqno = 0; |
- | |
1355 | 1898 | ||
1356 | if (ring->cleanup) |
1899 | if (ring->cleanup) |
1357 | ring->cleanup(ring); |
1900 | ring->cleanup(ring); |
1358 | 1901 | ||
1359 | i915_cmd_parser_fini_ring(ring); |
1902 | i915_cmd_parser_fini_ring(ring); |
- | 1903 | i915_gem_batch_pool_fini(&ring->batch_pool); |
|
1360 | 1904 | ||
1361 | if (ring->status_page.obj) { |
1905 | if (ring->status_page.obj) { |
1362 | kunmap(sg_page(ring->status_page.obj->pages->sgl)); |
1906 | kunmap(sg_page(ring->status_page.obj->pages->sgl)); |
1363 | ring->status_page.obj = NULL; |
1907 | ring->status_page.obj = NULL; |
1364 | } |
1908 | } |
- | 1909 | ||
- | 1910 | lrc_destroy_wa_ctx_obj(ring); |
|
1365 | } |
1911 | } |
1366 | 1912 | ||
1367 | static int logical_ring_init(struct drm_device *dev, struct intel_engine_cs *ring) |
1913 | static int logical_ring_init(struct drm_device *dev, struct intel_engine_cs *ring) |
1368 | { |
1914 | { |
1369 | int ret; |
1915 | int ret; |
1370 | 1916 | ||
1371 | /* Intentionally left blank. */ |
1917 | /* Intentionally left blank. */ |
1372 | ring->buffer = NULL; |
1918 | ring->buffer = NULL; |
1373 | 1919 | ||
1374 | ring->dev = dev; |
1920 | ring->dev = dev; |
1375 | INIT_LIST_HEAD(&ring->active_list); |
1921 | INIT_LIST_HEAD(&ring->active_list); |
1376 | INIT_LIST_HEAD(&ring->request_list); |
1922 | INIT_LIST_HEAD(&ring->request_list); |
- | 1923 | i915_gem_batch_pool_init(dev, &ring->batch_pool); |
|
1377 | init_waitqueue_head(&ring->irq_queue); |
1924 | init_waitqueue_head(&ring->irq_queue); |
1378 | 1925 | ||
1379 | INIT_LIST_HEAD(&ring->execlist_queue); |
1926 | INIT_LIST_HEAD(&ring->execlist_queue); |
1380 | INIT_LIST_HEAD(&ring->execlist_retired_req_list); |
1927 | INIT_LIST_HEAD(&ring->execlist_retired_req_list); |
1381 | spin_lock_init(&ring->execlist_lock); |
1928 | spin_lock_init(&ring->execlist_lock); |
1382 | ring->next_context_status_buffer = 0; |
- | |
1383 | 1929 | ||
1384 | ret = i915_cmd_parser_init_ring(ring); |
1930 | ret = i915_cmd_parser_init_ring(ring); |
1385 | if (ret) |
1931 | if (ret) |
1386 | return ret; |
1932 | return ret; |
1387 | - | ||
1388 | if (ring->init) { |
1933 | |
1389 | ret = ring->init(ring); |
1934 | ret = intel_lr_context_deferred_alloc(ring->default_context, ring); |
1390 | if (ret) |
1935 | if (ret) |
1391 | return ret; |
- | |
- | 1936 | return ret; |
|
- | 1937 | ||
- | 1938 | /* As this is the default context, always pin it */ |
|
- | 1939 | ret = intel_lr_context_do_pin( |
|
1392 | } |
1940 | ring, |
- | 1941 | ring->default_context->engine[ring->id].state, |
|
- | 1942 | ring->default_context->engine[ring->id].ringbuf); |
|
- | 1943 | if (ret) { |
|
- | 1944 | DRM_ERROR( |
|
- | 1945 | "Failed to pin and map ringbuffer %s: %d\n", |
|
- | 1946 | ring->name, ret); |
|
1393 | 1947 | return ret; |
|
1394 | ret = intel_lr_context_deferred_create(ring->default_context, ring); |
1948 | } |
1395 | 1949 | ||
1396 | return ret; |
1950 | return ret; |
1397 | } |
1951 | } |
1398 | 1952 | ||
1399 | static int logical_render_ring_init(struct drm_device *dev) |
1953 | static int logical_render_ring_init(struct drm_device *dev) |
1400 | { |
1954 | { |
1401 | struct drm_i915_private *dev_priv = dev->dev_private; |
1955 | struct drm_i915_private *dev_priv = dev->dev_private; |
1402 | struct intel_engine_cs *ring = &dev_priv->ring[RCS]; |
1956 | struct intel_engine_cs *ring = &dev_priv->ring[RCS]; |
- | 1957 | int ret; |
|
1403 | 1958 | ||
1404 | ring->name = "render ring"; |
1959 | ring->name = "render ring"; |
1405 | ring->id = RCS; |
1960 | ring->id = RCS; |
1406 | ring->mmio_base = RENDER_RING_BASE; |
1961 | ring->mmio_base = RENDER_RING_BASE; |
1407 | ring->irq_enable_mask = |
1962 | ring->irq_enable_mask = |
1408 | GT_RENDER_USER_INTERRUPT << GEN8_RCS_IRQ_SHIFT; |
1963 | GT_RENDER_USER_INTERRUPT << GEN8_RCS_IRQ_SHIFT; |
1409 | ring->irq_keep_mask = |
1964 | ring->irq_keep_mask = |
1410 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_RCS_IRQ_SHIFT; |
1965 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_RCS_IRQ_SHIFT; |
1411 | if (HAS_L3_DPF(dev)) |
1966 | if (HAS_L3_DPF(dev)) |
1412 | ring->irq_keep_mask |= GT_RENDER_L3_PARITY_ERROR_INTERRUPT; |
1967 | ring->irq_keep_mask |= GT_RENDER_L3_PARITY_ERROR_INTERRUPT; |
- | 1968 | ||
- | 1969 | if (INTEL_INFO(dev)->gen >= 9) |
|
- | 1970 | ring->init_hw = gen9_init_render_ring; |
|
1413 | 1971 | else |
|
1414 | ring->init = gen8_init_render_ring; |
1972 | ring->init_hw = gen8_init_render_ring; |
1415 | ring->init_context = intel_logical_ring_workarounds_emit; |
1973 | ring->init_context = gen8_init_rcs_context; |
- | 1974 | ring->cleanup = intel_fini_pipe_control; |
|
- | 1975 | if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) { |
|
- | 1976 | ring->get_seqno = bxt_a_get_seqno; |
|
- | 1977 | ring->set_seqno = bxt_a_set_seqno; |
|
1416 | ring->cleanup = intel_fini_pipe_control; |
1978 | } else { |
1417 | ring->get_seqno = gen8_get_seqno; |
1979 | ring->get_seqno = gen8_get_seqno; |
- | 1980 | ring->set_seqno = gen8_set_seqno; |
|
1418 | ring->set_seqno = gen8_set_seqno; |
1981 | } |
1419 | ring->emit_request = gen8_emit_request; |
1982 | ring->emit_request = gen8_emit_request; |
1420 | ring->emit_flush = gen8_emit_flush_render; |
1983 | ring->emit_flush = gen8_emit_flush_render; |
1421 | ring->irq_get = gen8_logical_ring_get_irq; |
1984 | ring->irq_get = gen8_logical_ring_get_irq; |
1422 | ring->irq_put = gen8_logical_ring_put_irq; |
1985 | ring->irq_put = gen8_logical_ring_put_irq; |
1423 | ring->emit_bb_start = gen8_emit_bb_start; |
1986 | ring->emit_bb_start = gen8_emit_bb_start; |
- | 1987 | ||
- | 1988 | ring->dev = dev; |
|
- | 1989 | ||
- | 1990 | ret = intel_init_pipe_control(ring); |
|
- | 1991 | if (ret) |
|
- | 1992 | return ret; |
|
- | 1993 | ||
- | 1994 | ret = intel_init_workaround_bb(ring); |
|
- | 1995 | if (ret) { |
|
- | 1996 | /* |
|
- | 1997 | * We continue even if we fail to initialize WA batch |
|
- | 1998 | * because we only expect rare glitches but nothing |
|
- | 1999 | * critical to prevent us from using GPU |
|
- | 2000 | */ |
|
- | 2001 | DRM_ERROR("WA batch buffer initialization failed: %d\n", |
|
- | 2002 | ret); |
|
- | 2003 | } |
|
1424 | 2004 | ||
- | 2005 | ret = logical_ring_init(dev, ring); |
|
- | 2006 | if (ret) { |
|
- | 2007 | lrc_destroy_wa_ctx_obj(ring); |
|
- | 2008 | } |
|
- | 2009 | ||
1425 | return logical_ring_init(dev, ring); |
2010 | return ret; |
1426 | } |
2011 | } |
1427 | 2012 | ||
1428 | static int logical_bsd_ring_init(struct drm_device *dev) |
2013 | static int logical_bsd_ring_init(struct drm_device *dev) |
1429 | { |
2014 | { |
1430 | struct drm_i915_private *dev_priv = dev->dev_private; |
2015 | struct drm_i915_private *dev_priv = dev->dev_private; |
1431 | struct intel_engine_cs *ring = &dev_priv->ring[VCS]; |
2016 | struct intel_engine_cs *ring = &dev_priv->ring[VCS]; |
1432 | 2017 | ||
1433 | ring->name = "bsd ring"; |
2018 | ring->name = "bsd ring"; |
1434 | ring->id = VCS; |
2019 | ring->id = VCS; |
1435 | ring->mmio_base = GEN6_BSD_RING_BASE; |
2020 | ring->mmio_base = GEN6_BSD_RING_BASE; |
1436 | ring->irq_enable_mask = |
2021 | ring->irq_enable_mask = |
1437 | GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; |
2022 | GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; |
1438 | ring->irq_keep_mask = |
2023 | ring->irq_keep_mask = |
1439 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; |
2024 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; |
1440 | 2025 | ||
- | 2026 | ring->init_hw = gen8_init_common_ring; |
|
- | 2027 | if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) { |
|
- | 2028 | ring->get_seqno = bxt_a_get_seqno; |
|
- | 2029 | ring->set_seqno = bxt_a_set_seqno; |
|
1441 | ring->init = gen8_init_common_ring; |
2030 | } else { |
1442 | ring->get_seqno = gen8_get_seqno; |
2031 | ring->get_seqno = gen8_get_seqno; |
- | 2032 | ring->set_seqno = gen8_set_seqno; |
|
1443 | ring->set_seqno = gen8_set_seqno; |
2033 | } |
1444 | ring->emit_request = gen8_emit_request; |
2034 | ring->emit_request = gen8_emit_request; |
1445 | ring->emit_flush = gen8_emit_flush; |
2035 | ring->emit_flush = gen8_emit_flush; |
1446 | ring->irq_get = gen8_logical_ring_get_irq; |
2036 | ring->irq_get = gen8_logical_ring_get_irq; |
1447 | ring->irq_put = gen8_logical_ring_put_irq; |
2037 | ring->irq_put = gen8_logical_ring_put_irq; |
1448 | ring->emit_bb_start = gen8_emit_bb_start; |
2038 | ring->emit_bb_start = gen8_emit_bb_start; |
1449 | 2039 | ||
1450 | return logical_ring_init(dev, ring); |
2040 | return logical_ring_init(dev, ring); |
1451 | } |
2041 | } |
1452 | 2042 | ||
1453 | static int logical_bsd2_ring_init(struct drm_device *dev) |
2043 | static int logical_bsd2_ring_init(struct drm_device *dev) |
1454 | { |
2044 | { |
1455 | struct drm_i915_private *dev_priv = dev->dev_private; |
2045 | struct drm_i915_private *dev_priv = dev->dev_private; |
1456 | struct intel_engine_cs *ring = &dev_priv->ring[VCS2]; |
2046 | struct intel_engine_cs *ring = &dev_priv->ring[VCS2]; |
1457 | 2047 | ||
1458 | ring->name = "bds2 ring"; |
2048 | ring->name = "bds2 ring"; |
1459 | ring->id = VCS2; |
2049 | ring->id = VCS2; |
1460 | ring->mmio_base = GEN8_BSD2_RING_BASE; |
2050 | ring->mmio_base = GEN8_BSD2_RING_BASE; |
1461 | ring->irq_enable_mask = |
2051 | ring->irq_enable_mask = |
1462 | GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; |
2052 | GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; |
1463 | ring->irq_keep_mask = |
2053 | ring->irq_keep_mask = |
1464 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; |
2054 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; |
1465 | 2055 | ||
1466 | ring->init = gen8_init_common_ring; |
2056 | ring->init_hw = gen8_init_common_ring; |
1467 | ring->get_seqno = gen8_get_seqno; |
2057 | ring->get_seqno = gen8_get_seqno; |
1468 | ring->set_seqno = gen8_set_seqno; |
2058 | ring->set_seqno = gen8_set_seqno; |
1469 | ring->emit_request = gen8_emit_request; |
2059 | ring->emit_request = gen8_emit_request; |
1470 | ring->emit_flush = gen8_emit_flush; |
2060 | ring->emit_flush = gen8_emit_flush; |
1471 | ring->irq_get = gen8_logical_ring_get_irq; |
2061 | ring->irq_get = gen8_logical_ring_get_irq; |
1472 | ring->irq_put = gen8_logical_ring_put_irq; |
2062 | ring->irq_put = gen8_logical_ring_put_irq; |
1473 | ring->emit_bb_start = gen8_emit_bb_start; |
2063 | ring->emit_bb_start = gen8_emit_bb_start; |
1474 | 2064 | ||
1475 | return logical_ring_init(dev, ring); |
2065 | return logical_ring_init(dev, ring); |
1476 | } |
2066 | } |
1477 | 2067 | ||
1478 | static int logical_blt_ring_init(struct drm_device *dev) |
2068 | static int logical_blt_ring_init(struct drm_device *dev) |
1479 | { |
2069 | { |
1480 | struct drm_i915_private *dev_priv = dev->dev_private; |
2070 | struct drm_i915_private *dev_priv = dev->dev_private; |
1481 | struct intel_engine_cs *ring = &dev_priv->ring[BCS]; |
2071 | struct intel_engine_cs *ring = &dev_priv->ring[BCS]; |
1482 | 2072 | ||
1483 | ring->name = "blitter ring"; |
2073 | ring->name = "blitter ring"; |
1484 | ring->id = BCS; |
2074 | ring->id = BCS; |
1485 | ring->mmio_base = BLT_RING_BASE; |
2075 | ring->mmio_base = BLT_RING_BASE; |
1486 | ring->irq_enable_mask = |
2076 | ring->irq_enable_mask = |
1487 | GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT; |
2077 | GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT; |
1488 | ring->irq_keep_mask = |
2078 | ring->irq_keep_mask = |
1489 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_BCS_IRQ_SHIFT; |
2079 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_BCS_IRQ_SHIFT; |
1490 | 2080 | ||
- | 2081 | ring->init_hw = gen8_init_common_ring; |
|
- | 2082 | if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) { |
|
- | 2083 | ring->get_seqno = bxt_a_get_seqno; |
|
- | 2084 | ring->set_seqno = bxt_a_set_seqno; |
|
1491 | ring->init = gen8_init_common_ring; |
2085 | } else { |
1492 | ring->get_seqno = gen8_get_seqno; |
2086 | ring->get_seqno = gen8_get_seqno; |
- | 2087 | ring->set_seqno = gen8_set_seqno; |
|
1493 | ring->set_seqno = gen8_set_seqno; |
2088 | } |
1494 | ring->emit_request = gen8_emit_request; |
2089 | ring->emit_request = gen8_emit_request; |
1495 | ring->emit_flush = gen8_emit_flush; |
2090 | ring->emit_flush = gen8_emit_flush; |
1496 | ring->irq_get = gen8_logical_ring_get_irq; |
2091 | ring->irq_get = gen8_logical_ring_get_irq; |
1497 | ring->irq_put = gen8_logical_ring_put_irq; |
2092 | ring->irq_put = gen8_logical_ring_put_irq; |
1498 | ring->emit_bb_start = gen8_emit_bb_start; |
2093 | ring->emit_bb_start = gen8_emit_bb_start; |
1499 | 2094 | ||
1500 | return logical_ring_init(dev, ring); |
2095 | return logical_ring_init(dev, ring); |
1501 | } |
2096 | } |
1502 | 2097 | ||
1503 | static int logical_vebox_ring_init(struct drm_device *dev) |
2098 | static int logical_vebox_ring_init(struct drm_device *dev) |
1504 | { |
2099 | { |
1505 | struct drm_i915_private *dev_priv = dev->dev_private; |
2100 | struct drm_i915_private *dev_priv = dev->dev_private; |
1506 | struct intel_engine_cs *ring = &dev_priv->ring[VECS]; |
2101 | struct intel_engine_cs *ring = &dev_priv->ring[VECS]; |
1507 | 2102 | ||
1508 | ring->name = "video enhancement ring"; |
2103 | ring->name = "video enhancement ring"; |
1509 | ring->id = VECS; |
2104 | ring->id = VECS; |
1510 | ring->mmio_base = VEBOX_RING_BASE; |
2105 | ring->mmio_base = VEBOX_RING_BASE; |
1511 | ring->irq_enable_mask = |
2106 | ring->irq_enable_mask = |
1512 | GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT; |
2107 | GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT; |
1513 | ring->irq_keep_mask = |
2108 | ring->irq_keep_mask = |
1514 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VECS_IRQ_SHIFT; |
2109 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VECS_IRQ_SHIFT; |
1515 | 2110 | ||
- | 2111 | ring->init_hw = gen8_init_common_ring; |
|
- | 2112 | if (IS_BROXTON(dev) && INTEL_REVID(dev) < BXT_REVID_B0) { |
|
- | 2113 | ring->get_seqno = bxt_a_get_seqno; |
|
- | 2114 | ring->set_seqno = bxt_a_set_seqno; |
|
1516 | ring->init = gen8_init_common_ring; |
2115 | } else { |
1517 | ring->get_seqno = gen8_get_seqno; |
2116 | ring->get_seqno = gen8_get_seqno; |
- | 2117 | ring->set_seqno = gen8_set_seqno; |
|
1518 | ring->set_seqno = gen8_set_seqno; |
2118 | } |
1519 | ring->emit_request = gen8_emit_request; |
2119 | ring->emit_request = gen8_emit_request; |
1520 | ring->emit_flush = gen8_emit_flush; |
2120 | ring->emit_flush = gen8_emit_flush; |
1521 | ring->irq_get = gen8_logical_ring_get_irq; |
2121 | ring->irq_get = gen8_logical_ring_get_irq; |
1522 | ring->irq_put = gen8_logical_ring_put_irq; |
2122 | ring->irq_put = gen8_logical_ring_put_irq; |
1523 | ring->emit_bb_start = gen8_emit_bb_start; |
2123 | ring->emit_bb_start = gen8_emit_bb_start; |
1524 | 2124 | ||
1525 | return logical_ring_init(dev, ring); |
2125 | return logical_ring_init(dev, ring); |
1526 | } |
2126 | } |
1527 | 2127 | ||
1528 | /** |
2128 | /** |
1529 | * intel_logical_rings_init() - allocate, populate and init the Engine Command Streamers |
2129 | * intel_logical_rings_init() - allocate, populate and init the Engine Command Streamers |
1530 | * @dev: DRM device. |
2130 | * @dev: DRM device. |
1531 | * |
2131 | * |
1532 | * This function inits the engines for an Execlists submission style (the equivalent in the |
2132 | * This function inits the engines for an Execlists submission style (the equivalent in the |
1533 | * legacy ringbuffer submission world would be i915_gem_init_rings). It does it only for |
2133 | * legacy ringbuffer submission world would be i915_gem_init_rings). It does it only for |
1534 | * those engines that are present in the hardware. |
2134 | * those engines that are present in the hardware. |
1535 | * |
2135 | * |
1536 | * Return: non-zero if the initialization failed. |
2136 | * Return: non-zero if the initialization failed. |
1537 | */ |
2137 | */ |
1538 | int intel_logical_rings_init(struct drm_device *dev) |
2138 | int intel_logical_rings_init(struct drm_device *dev) |
1539 | { |
2139 | { |
1540 | struct drm_i915_private *dev_priv = dev->dev_private; |
2140 | struct drm_i915_private *dev_priv = dev->dev_private; |
1541 | int ret; |
2141 | int ret; |
1542 | 2142 | ||
1543 | ret = logical_render_ring_init(dev); |
2143 | ret = logical_render_ring_init(dev); |
1544 | if (ret) |
2144 | if (ret) |
1545 | return ret; |
2145 | return ret; |
1546 | 2146 | ||
1547 | if (HAS_BSD(dev)) { |
2147 | if (HAS_BSD(dev)) { |
1548 | ret = logical_bsd_ring_init(dev); |
2148 | ret = logical_bsd_ring_init(dev); |
1549 | if (ret) |
2149 | if (ret) |
1550 | goto cleanup_render_ring; |
2150 | goto cleanup_render_ring; |
1551 | } |
2151 | } |
1552 | 2152 | ||
1553 | if (HAS_BLT(dev)) { |
2153 | if (HAS_BLT(dev)) { |
1554 | ret = logical_blt_ring_init(dev); |
2154 | ret = logical_blt_ring_init(dev); |
1555 | if (ret) |
2155 | if (ret) |
1556 | goto cleanup_bsd_ring; |
2156 | goto cleanup_bsd_ring; |
1557 | } |
2157 | } |
1558 | 2158 | ||
1559 | if (HAS_VEBOX(dev)) { |
2159 | if (HAS_VEBOX(dev)) { |
1560 | ret = logical_vebox_ring_init(dev); |
2160 | ret = logical_vebox_ring_init(dev); |
1561 | if (ret) |
2161 | if (ret) |
1562 | goto cleanup_blt_ring; |
2162 | goto cleanup_blt_ring; |
1563 | } |
2163 | } |
1564 | 2164 | ||
1565 | if (HAS_BSD2(dev)) { |
2165 | if (HAS_BSD2(dev)) { |
1566 | ret = logical_bsd2_ring_init(dev); |
2166 | ret = logical_bsd2_ring_init(dev); |
1567 | if (ret) |
2167 | if (ret) |
1568 | goto cleanup_vebox_ring; |
2168 | goto cleanup_vebox_ring; |
1569 | } |
2169 | } |
1570 | - | ||
1571 | ret = i915_gem_set_seqno(dev, ((u32)~0 - 0x1000)); |
- | |
1572 | if (ret) |
- | |
1573 | goto cleanup_bsd2_ring; |
- | |
1574 | 2170 | ||
1575 | return 0; |
- | |
1576 | - | ||
1577 | cleanup_bsd2_ring: |
2171 | return 0; |
1578 | intel_logical_ring_cleanup(&dev_priv->ring[VCS2]); |
2172 | |
1579 | cleanup_vebox_ring: |
2173 | cleanup_vebox_ring: |
1580 | intel_logical_ring_cleanup(&dev_priv->ring[VECS]); |
2174 | intel_logical_ring_cleanup(&dev_priv->ring[VECS]); |
1581 | cleanup_blt_ring: |
2175 | cleanup_blt_ring: |
1582 | intel_logical_ring_cleanup(&dev_priv->ring[BCS]); |
2176 | intel_logical_ring_cleanup(&dev_priv->ring[BCS]); |
1583 | cleanup_bsd_ring: |
2177 | cleanup_bsd_ring: |
1584 | intel_logical_ring_cleanup(&dev_priv->ring[VCS]); |
2178 | intel_logical_ring_cleanup(&dev_priv->ring[VCS]); |
1585 | cleanup_render_ring: |
2179 | cleanup_render_ring: |
1586 | intel_logical_ring_cleanup(&dev_priv->ring[RCS]); |
2180 | intel_logical_ring_cleanup(&dev_priv->ring[RCS]); |
1587 | 2181 | ||
1588 | return ret; |
2182 | return ret; |
1589 | } |
2183 | } |
1590 | 2184 | ||
1591 | int intel_lr_context_render_state_init(struct intel_engine_cs *ring, |
2185 | static u32 |
1592 | struct intel_context *ctx) |
2186 | make_rpcs(struct drm_device *dev) |
1593 | { |
- | |
1594 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
- | |
1595 | struct render_state so; |
- | |
1596 | struct drm_i915_file_private *file_priv = ctx->file_priv; |
- | |
1597 | struct drm_file *file = file_priv ? file_priv->file : NULL; |
2187 | { |
- | 2188 | u32 rpcs = 0; |
|
1598 | int ret; |
2189 | |
1599 | - | ||
1600 | ret = i915_gem_render_state_prepare(ring, &so); |
2190 | /* |
1601 | if (ret) |
2191 | * No explicit RPCS request is needed to ensure full |
1602 | return ret; |
2192 | * slice/subslice/EU enablement prior to Gen9. |
1603 | 2193 | */ |
|
- | 2194 | if (INTEL_INFO(dev)->gen < 9) |
|
1604 | if (so.rodata == NULL) |
2195 | return 0; |
- | 2196 | ||
- | 2197 | /* |
|
1605 | return 0; |
2198 | * Starting in Gen9, render power gating can leave |
- | 2199 | * slice/subslice/EU in a partially enabled state. We |
|
- | 2200 | * must make an explicit request through RPCS for full |
|
- | 2201 | * enablement. |
|
- | 2202 | */ |
|
1606 | 2203 | if (INTEL_INFO(dev)->has_slice_pg) { |
|
1607 | ret = ring->emit_bb_start(ringbuf, |
2204 | rpcs |= GEN8_RPCS_S_CNT_ENABLE; |
1608 | so.ggtt_offset, |
2205 | rpcs |= INTEL_INFO(dev)->slice_total << |
- | 2206 | GEN8_RPCS_S_CNT_SHIFT; |
|
- | 2207 | rpcs |= GEN8_RPCS_ENABLE; |
|
1609 | I915_DISPATCH_SECURE); |
2208 | } |
- | 2209 | ||
- | 2210 | if (INTEL_INFO(dev)->has_subslice_pg) { |
|
- | 2211 | rpcs |= GEN8_RPCS_SS_CNT_ENABLE; |
|
- | 2212 | rpcs |= INTEL_INFO(dev)->subslice_per_slice << |
|
1610 | if (ret) |
2213 | GEN8_RPCS_SS_CNT_SHIFT; |
- | 2214 | rpcs |= GEN8_RPCS_ENABLE; |
|
1611 | goto out; |
2215 | } |
- | 2216 | ||
1612 | 2217 | if (INTEL_INFO(dev)->has_eu_pg) { |
|
1613 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), ring); |
2218 | rpcs |= INTEL_INFO(dev)->eu_per_subslice << |
1614 | - | ||
- | 2219 | GEN8_RPCS_EU_MIN_SHIFT; |
|
1615 | ret = __i915_add_request(ring, file, so.obj, NULL); |
2220 | rpcs |= INTEL_INFO(dev)->eu_per_subslice << |
1616 | /* intel_logical_ring_add_request moves object to inactive if it |
2221 | GEN8_RPCS_EU_MAX_SHIFT; |
1617 | * fails */ |
2222 | rpcs |= GEN8_RPCS_ENABLE; |
1618 | out: |
2223 | } |
1619 | i915_gem_render_state_fini(&so); |
2224 | |
1620 | return ret; |
2225 | return rpcs; |
1621 | } |
2226 | } |
1622 | 2227 | ||
1623 | static int |
2228 | static int |
1624 | populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_obj, |
2229 | populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_obj, |
1625 | struct intel_engine_cs *ring, struct intel_ringbuffer *ringbuf) |
2230 | struct intel_engine_cs *ring, struct intel_ringbuffer *ringbuf) |
1626 | { |
2231 | { |
1627 | struct drm_device *dev = ring->dev; |
2232 | struct drm_device *dev = ring->dev; |
1628 | struct drm_i915_private *dev_priv = dev->dev_private; |
2233 | struct drm_i915_private *dev_priv = dev->dev_private; |
1629 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; |
2234 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; |
1630 | struct page *page; |
2235 | struct page *page; |
1631 | uint32_t *reg_state; |
2236 | uint32_t *reg_state; |
1632 | int ret; |
2237 | int ret; |
1633 | 2238 | ||
1634 | if (!ppgtt) |
2239 | if (!ppgtt) |
1635 | ppgtt = dev_priv->mm.aliasing_ppgtt; |
2240 | ppgtt = dev_priv->mm.aliasing_ppgtt; |
1636 | 2241 | ||
1637 | ret = i915_gem_object_set_to_cpu_domain(ctx_obj, true); |
2242 | ret = i915_gem_object_set_to_cpu_domain(ctx_obj, true); |
1638 | if (ret) { |
2243 | if (ret) { |
1639 | DRM_DEBUG_DRIVER("Could not set to CPU domain\n"); |
2244 | DRM_DEBUG_DRIVER("Could not set to CPU domain\n"); |
1640 | return ret; |
2245 | return ret; |
1641 | } |
2246 | } |
1642 | 2247 | ||
1643 | ret = i915_gem_object_get_pages(ctx_obj); |
2248 | ret = i915_gem_object_get_pages(ctx_obj); |
1644 | if (ret) { |
2249 | if (ret) { |
1645 | DRM_DEBUG_DRIVER("Could not get object pages\n"); |
2250 | DRM_DEBUG_DRIVER("Could not get object pages\n"); |
1646 | return ret; |
2251 | return ret; |
1647 | } |
2252 | } |
1648 | 2253 | ||
1649 | i915_gem_object_pin_pages(ctx_obj); |
2254 | i915_gem_object_pin_pages(ctx_obj); |
1650 | 2255 | ||
1651 | /* The second page of the context object contains some fields which must |
2256 | /* The second page of the context object contains some fields which must |
1652 | * be set up prior to the first execution. */ |
2257 | * be set up prior to the first execution. */ |
1653 | page = i915_gem_object_get_page(ctx_obj, 1); |
2258 | page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); |
1654 | reg_state = kmap_atomic(page); |
2259 | reg_state = kmap_atomic(page); |
1655 | 2260 | ||
1656 | /* A context is actually a big batch buffer with several MI_LOAD_REGISTER_IMM |
2261 | /* A context is actually a big batch buffer with several MI_LOAD_REGISTER_IMM |
1657 | * commands followed by (reg, value) pairs. The values we are setting here are |
2262 | * commands followed by (reg, value) pairs. The values we are setting here are |
1658 | * only for the first context restore: on a subsequent save, the GPU will |
2263 | * only for the first context restore: on a subsequent save, the GPU will |
1659 | * recreate this batchbuffer with new values (including all the missing |
2264 | * recreate this batchbuffer with new values (including all the missing |
1660 | * MI_LOAD_REGISTER_IMM commands that we are not initializing here). */ |
2265 | * MI_LOAD_REGISTER_IMM commands that we are not initializing here). */ |
1661 | if (ring->id == RCS) |
2266 | if (ring->id == RCS) |
1662 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(14); |
2267 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(14); |
1663 | else |
2268 | else |
1664 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(11); |
2269 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(11); |
1665 | reg_state[CTX_LRI_HEADER_0] |= MI_LRI_FORCE_POSTED; |
2270 | reg_state[CTX_LRI_HEADER_0] |= MI_LRI_FORCE_POSTED; |
1666 | reg_state[CTX_CONTEXT_CONTROL] = RING_CONTEXT_CONTROL(ring); |
2271 | reg_state[CTX_CONTEXT_CONTROL] = RING_CONTEXT_CONTROL(ring); |
1667 | reg_state[CTX_CONTEXT_CONTROL+1] = |
2272 | reg_state[CTX_CONTEXT_CONTROL+1] = |
1668 | _MASKED_BIT_ENABLE((1<<3) | MI_RESTORE_INHIBIT); |
2273 | _MASKED_BIT_ENABLE(CTX_CTRL_INHIBIT_SYN_CTX_SWITCH | |
- | 2274 | CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT | |
|
- | 2275 | CTX_CTRL_RS_CTX_ENABLE); |
|
1669 | reg_state[CTX_RING_HEAD] = RING_HEAD(ring->mmio_base); |
2276 | reg_state[CTX_RING_HEAD] = RING_HEAD(ring->mmio_base); |
1670 | reg_state[CTX_RING_HEAD+1] = 0; |
2277 | reg_state[CTX_RING_HEAD+1] = 0; |
1671 | reg_state[CTX_RING_TAIL] = RING_TAIL(ring->mmio_base); |
2278 | reg_state[CTX_RING_TAIL] = RING_TAIL(ring->mmio_base); |
1672 | reg_state[CTX_RING_TAIL+1] = 0; |
2279 | reg_state[CTX_RING_TAIL+1] = 0; |
1673 | reg_state[CTX_RING_BUFFER_START] = RING_START(ring->mmio_base); |
2280 | reg_state[CTX_RING_BUFFER_START] = RING_START(ring->mmio_base); |
1674 | /* Ring buffer start address is not known until the buffer is pinned. |
2281 | /* Ring buffer start address is not known until the buffer is pinned. |
1675 | * It is written to the context image in execlists_update_context() |
2282 | * It is written to the context image in execlists_update_context() |
1676 | */ |
2283 | */ |
1677 | reg_state[CTX_RING_BUFFER_CONTROL] = RING_CTL(ring->mmio_base); |
2284 | reg_state[CTX_RING_BUFFER_CONTROL] = RING_CTL(ring->mmio_base); |
1678 | reg_state[CTX_RING_BUFFER_CONTROL+1] = |
2285 | reg_state[CTX_RING_BUFFER_CONTROL+1] = |
1679 | ((ringbuf->size - PAGE_SIZE) & RING_NR_PAGES) | RING_VALID; |
2286 | ((ringbuf->size - PAGE_SIZE) & RING_NR_PAGES) | RING_VALID; |
1680 | reg_state[CTX_BB_HEAD_U] = ring->mmio_base + 0x168; |
2287 | reg_state[CTX_BB_HEAD_U] = ring->mmio_base + 0x168; |
1681 | reg_state[CTX_BB_HEAD_U+1] = 0; |
2288 | reg_state[CTX_BB_HEAD_U+1] = 0; |
1682 | reg_state[CTX_BB_HEAD_L] = ring->mmio_base + 0x140; |
2289 | reg_state[CTX_BB_HEAD_L] = ring->mmio_base + 0x140; |
1683 | reg_state[CTX_BB_HEAD_L+1] = 0; |
2290 | reg_state[CTX_BB_HEAD_L+1] = 0; |
1684 | reg_state[CTX_BB_STATE] = ring->mmio_base + 0x110; |
2291 | reg_state[CTX_BB_STATE] = ring->mmio_base + 0x110; |
1685 | reg_state[CTX_BB_STATE+1] = (1<<5); |
2292 | reg_state[CTX_BB_STATE+1] = (1<<5); |
1686 | reg_state[CTX_SECOND_BB_HEAD_U] = ring->mmio_base + 0x11c; |
2293 | reg_state[CTX_SECOND_BB_HEAD_U] = ring->mmio_base + 0x11c; |
1687 | reg_state[CTX_SECOND_BB_HEAD_U+1] = 0; |
2294 | reg_state[CTX_SECOND_BB_HEAD_U+1] = 0; |
1688 | reg_state[CTX_SECOND_BB_HEAD_L] = ring->mmio_base + 0x114; |
2295 | reg_state[CTX_SECOND_BB_HEAD_L] = ring->mmio_base + 0x114; |
1689 | reg_state[CTX_SECOND_BB_HEAD_L+1] = 0; |
2296 | reg_state[CTX_SECOND_BB_HEAD_L+1] = 0; |
1690 | reg_state[CTX_SECOND_BB_STATE] = ring->mmio_base + 0x118; |
2297 | reg_state[CTX_SECOND_BB_STATE] = ring->mmio_base + 0x118; |
1691 | reg_state[CTX_SECOND_BB_STATE+1] = 0; |
2298 | reg_state[CTX_SECOND_BB_STATE+1] = 0; |
1692 | if (ring->id == RCS) { |
2299 | if (ring->id == RCS) { |
1693 | /* TODO: according to BSpec, the register state context |
- | |
1694 | * for CHV does not have these. OTOH, these registers do |
- | |
1695 | * exist in CHV. I'm waiting for a clarification */ |
- | |
1696 | reg_state[CTX_BB_PER_CTX_PTR] = ring->mmio_base + 0x1c0; |
2300 | reg_state[CTX_BB_PER_CTX_PTR] = ring->mmio_base + 0x1c0; |
1697 | reg_state[CTX_BB_PER_CTX_PTR+1] = 0; |
2301 | reg_state[CTX_BB_PER_CTX_PTR+1] = 0; |
1698 | reg_state[CTX_RCS_INDIRECT_CTX] = ring->mmio_base + 0x1c4; |
2302 | reg_state[CTX_RCS_INDIRECT_CTX] = ring->mmio_base + 0x1c4; |
1699 | reg_state[CTX_RCS_INDIRECT_CTX+1] = 0; |
2303 | reg_state[CTX_RCS_INDIRECT_CTX+1] = 0; |
1700 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET] = ring->mmio_base + 0x1c8; |
2304 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET] = ring->mmio_base + 0x1c8; |
1701 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET+1] = 0; |
2305 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET+1] = 0; |
- | 2306 | if (ring->wa_ctx.obj) { |
|
- | 2307 | struct i915_ctx_workarounds *wa_ctx = &ring->wa_ctx; |
|
- | 2308 | uint32_t ggtt_offset = i915_gem_obj_ggtt_offset(wa_ctx->obj); |
|
- | 2309 | ||
- | 2310 | reg_state[CTX_RCS_INDIRECT_CTX+1] = |
|
- | 2311 | (ggtt_offset + wa_ctx->indirect_ctx.offset * sizeof(uint32_t)) | |
|
- | 2312 | (wa_ctx->indirect_ctx.size / CACHELINE_DWORDS); |
|
- | 2313 | ||
- | 2314 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET+1] = |
|
- | 2315 | CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT << 6; |
|
- | 2316 | ||
- | 2317 | reg_state[CTX_BB_PER_CTX_PTR+1] = |
|
- | 2318 | (ggtt_offset + wa_ctx->per_ctx.offset * sizeof(uint32_t)) | |
|
- | 2319 | 0x01; |
|
- | 2320 | } |
|
1702 | } |
2321 | } |
1703 | reg_state[CTX_LRI_HEADER_1] = MI_LOAD_REGISTER_IMM(9); |
2322 | reg_state[CTX_LRI_HEADER_1] = MI_LOAD_REGISTER_IMM(9); |
1704 | reg_state[CTX_LRI_HEADER_1] |= MI_LRI_FORCE_POSTED; |
2323 | reg_state[CTX_LRI_HEADER_1] |= MI_LRI_FORCE_POSTED; |
1705 | reg_state[CTX_CTX_TIMESTAMP] = ring->mmio_base + 0x3a8; |
2324 | reg_state[CTX_CTX_TIMESTAMP] = ring->mmio_base + 0x3a8; |
1706 | reg_state[CTX_CTX_TIMESTAMP+1] = 0; |
2325 | reg_state[CTX_CTX_TIMESTAMP+1] = 0; |
1707 | reg_state[CTX_PDP3_UDW] = GEN8_RING_PDP_UDW(ring, 3); |
2326 | reg_state[CTX_PDP3_UDW] = GEN8_RING_PDP_UDW(ring, 3); |
1708 | reg_state[CTX_PDP3_LDW] = GEN8_RING_PDP_LDW(ring, 3); |
2327 | reg_state[CTX_PDP3_LDW] = GEN8_RING_PDP_LDW(ring, 3); |
1709 | reg_state[CTX_PDP2_UDW] = GEN8_RING_PDP_UDW(ring, 2); |
2328 | reg_state[CTX_PDP2_UDW] = GEN8_RING_PDP_UDW(ring, 2); |
1710 | reg_state[CTX_PDP2_LDW] = GEN8_RING_PDP_LDW(ring, 2); |
2329 | reg_state[CTX_PDP2_LDW] = GEN8_RING_PDP_LDW(ring, 2); |
1711 | reg_state[CTX_PDP1_UDW] = GEN8_RING_PDP_UDW(ring, 1); |
2330 | reg_state[CTX_PDP1_UDW] = GEN8_RING_PDP_UDW(ring, 1); |
1712 | reg_state[CTX_PDP1_LDW] = GEN8_RING_PDP_LDW(ring, 1); |
2331 | reg_state[CTX_PDP1_LDW] = GEN8_RING_PDP_LDW(ring, 1); |
1713 | reg_state[CTX_PDP0_UDW] = GEN8_RING_PDP_UDW(ring, 0); |
2332 | reg_state[CTX_PDP0_UDW] = GEN8_RING_PDP_UDW(ring, 0); |
1714 | reg_state[CTX_PDP0_LDW] = GEN8_RING_PDP_LDW(ring, 0); |
2333 | reg_state[CTX_PDP0_LDW] = GEN8_RING_PDP_LDW(ring, 0); |
- | 2334 | ||
1715 | reg_state[CTX_PDP3_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[3]); |
2335 | if (USES_FULL_48BIT_PPGTT(ppgtt->base.dev)) { |
- | 2336 | /* 64b PPGTT (48bit canonical) |
|
- | 2337 | * PDP0_DESCRIPTOR contains the base address to PML4 and |
|
- | 2338 | * other PDP Descriptors are ignored. |
|
- | 2339 | */ |
|
- | 2340 | ASSIGN_CTX_PML4(ppgtt, reg_state); |
|
- | 2341 | } else { |
|
- | 2342 | /* 32b PPGTT |
|
1716 | reg_state[CTX_PDP3_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[3]); |
2343 | * PDP*_DESCRIPTOR contains the base address of space supported. |
1717 | reg_state[CTX_PDP2_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[2]); |
2344 | * With dynamic page allocation, PDPs may not be allocated at |
1718 | reg_state[CTX_PDP2_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[2]); |
2345 | * this point. Point the unallocated PDPs to the scratch page |
- | 2346 | */ |
|
1719 | reg_state[CTX_PDP1_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[1]); |
2347 | ASSIGN_CTX_PDP(ppgtt, reg_state, 3); |
1720 | reg_state[CTX_PDP1_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[1]); |
2348 | ASSIGN_CTX_PDP(ppgtt, reg_state, 2); |
1721 | reg_state[CTX_PDP0_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[0]); |
2349 | ASSIGN_CTX_PDP(ppgtt, reg_state, 1); |
1722 | reg_state[CTX_PDP0_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[0]); |
2350 | ASSIGN_CTX_PDP(ppgtt, reg_state, 0); |
- | 2351 | } |
|
- | 2352 | ||
1723 | if (ring->id == RCS) { |
2353 | if (ring->id == RCS) { |
1724 | reg_state[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1); |
2354 | reg_state[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1); |
1725 | reg_state[CTX_R_PWR_CLK_STATE] = 0x20c8; |
2355 | reg_state[CTX_R_PWR_CLK_STATE] = GEN8_R_PWR_CLK_STATE; |
1726 | reg_state[CTX_R_PWR_CLK_STATE+1] = 0; |
2356 | reg_state[CTX_R_PWR_CLK_STATE+1] = make_rpcs(dev); |
1727 | } |
2357 | } |
1728 | 2358 | ||
1729 | kunmap_atomic(reg_state); |
2359 | kunmap_atomic(reg_state); |
1730 | 2360 | ||
1731 | ctx_obj->dirty = 1; |
2361 | ctx_obj->dirty = 1; |
1732 | // set_page_dirty(page); |
- | |
1733 | i915_gem_object_unpin_pages(ctx_obj); |
2362 | i915_gem_object_unpin_pages(ctx_obj); |
1734 | 2363 | ||
1735 | return 0; |
2364 | return 0; |
1736 | } |
2365 | } |
1737 | 2366 | ||
1738 | /** |
2367 | /** |
1739 | * intel_lr_context_free() - free the LRC specific bits of a context |
2368 | * intel_lr_context_free() - free the LRC specific bits of a context |
1740 | * @ctx: the LR context to free. |
2369 | * @ctx: the LR context to free. |
1741 | * |
2370 | * |
1742 | * The real context freeing is done in i915_gem_context_free: this only |
2371 | * The real context freeing is done in i915_gem_context_free: this only |
1743 | * takes care of the bits that are LRC related: the per-engine backing |
2372 | * takes care of the bits that are LRC related: the per-engine backing |
1744 | * objects and the logical ringbuffer. |
2373 | * objects and the logical ringbuffer. |
1745 | */ |
2374 | */ |
1746 | void intel_lr_context_free(struct intel_context *ctx) |
2375 | void intel_lr_context_free(struct intel_context *ctx) |
1747 | { |
2376 | { |
1748 | int i; |
2377 | int i; |
1749 | 2378 | ||
1750 | for (i = 0; i < I915_NUM_RINGS; i++) { |
2379 | for (i = 0; i < I915_NUM_RINGS; i++) { |
1751 | struct drm_i915_gem_object *ctx_obj = ctx->engine[i].state; |
2380 | struct drm_i915_gem_object *ctx_obj = ctx->engine[i].state; |
1752 | 2381 | ||
1753 | if (ctx_obj) { |
2382 | if (ctx_obj) { |
1754 | struct intel_ringbuffer *ringbuf = |
2383 | struct intel_ringbuffer *ringbuf = |
1755 | ctx->engine[i].ringbuf; |
2384 | ctx->engine[i].ringbuf; |
1756 | struct intel_engine_cs *ring = ringbuf->ring; |
2385 | struct intel_engine_cs *ring = ringbuf->ring; |
1757 | 2386 | ||
1758 | if (ctx == ring->default_context) { |
2387 | if (ctx == ring->default_context) { |
1759 | intel_unpin_ringbuffer_obj(ringbuf); |
2388 | intel_unpin_ringbuffer_obj(ringbuf); |
1760 | i915_gem_object_ggtt_unpin(ctx_obj); |
2389 | i915_gem_object_ggtt_unpin(ctx_obj); |
1761 | } |
2390 | } |
1762 | intel_destroy_ringbuffer_obj(ringbuf); |
2391 | WARN_ON(ctx->engine[ring->id].pin_count); |
1763 | kfree(ringbuf); |
2392 | intel_ringbuffer_free(ringbuf); |
1764 | drm_gem_object_unreference(&ctx_obj->base); |
2393 | drm_gem_object_unreference(&ctx_obj->base); |
1765 | } |
2394 | } |
1766 | } |
2395 | } |
1767 | } |
2396 | } |
1768 | 2397 | ||
1769 | static uint32_t get_lr_context_size(struct intel_engine_cs *ring) |
2398 | static uint32_t get_lr_context_size(struct intel_engine_cs *ring) |
1770 | { |
2399 | { |
1771 | int ret = 0; |
2400 | int ret = 0; |
1772 | 2401 | ||
1773 | WARN_ON(INTEL_INFO(ring->dev)->gen < 8); |
2402 | WARN_ON(INTEL_INFO(ring->dev)->gen < 8); |
1774 | 2403 | ||
1775 | switch (ring->id) { |
2404 | switch (ring->id) { |
1776 | case RCS: |
2405 | case RCS: |
1777 | if (INTEL_INFO(ring->dev)->gen >= 9) |
2406 | if (INTEL_INFO(ring->dev)->gen >= 9) |
1778 | ret = GEN9_LR_CONTEXT_RENDER_SIZE; |
2407 | ret = GEN9_LR_CONTEXT_RENDER_SIZE; |
1779 | else |
2408 | else |
1780 | ret = GEN8_LR_CONTEXT_RENDER_SIZE; |
2409 | ret = GEN8_LR_CONTEXT_RENDER_SIZE; |
1781 | break; |
2410 | break; |
1782 | case VCS: |
2411 | case VCS: |
1783 | case BCS: |
2412 | case BCS: |
1784 | case VECS: |
2413 | case VECS: |
1785 | case VCS2: |
2414 | case VCS2: |
1786 | ret = GEN8_LR_CONTEXT_OTHER_SIZE; |
2415 | ret = GEN8_LR_CONTEXT_OTHER_SIZE; |
1787 | break; |
2416 | break; |
1788 | } |
2417 | } |
1789 | 2418 | ||
1790 | return ret; |
2419 | return ret; |
1791 | } |
2420 | } |
1792 | 2421 | ||
1793 | static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, |
2422 | static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, |
1794 | struct drm_i915_gem_object *default_ctx_obj) |
2423 | struct drm_i915_gem_object *default_ctx_obj) |
1795 | { |
2424 | { |
1796 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
2425 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
- | 2426 | struct page *page; |
|
1797 | 2427 | ||
1798 | /* The status page is offset 0 from the default context object |
- | |
1799 | * in LRC mode. */ |
2428 | /* The HWSP is part of the default context object in LRC mode. */ |
1800 | ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj); |
2429 | ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj) |
1801 | ring->status_page.page_addr = |
2430 | + LRC_PPHWSP_PN * PAGE_SIZE; |
- | 2431 | page = i915_gem_object_get_page(default_ctx_obj, LRC_PPHWSP_PN); |
|
1802 | kmap(sg_page(default_ctx_obj->pages->sgl)); |
2432 | ring->status_page.page_addr = kmap(page); |
1803 | ring->status_page.obj = default_ctx_obj; |
2433 | ring->status_page.obj = default_ctx_obj; |
1804 | 2434 | ||
1805 | I915_WRITE(RING_HWS_PGA(ring->mmio_base), |
2435 | I915_WRITE(RING_HWS_PGA(ring->mmio_base), |
1806 | (u32)ring->status_page.gfx_addr); |
2436 | (u32)ring->status_page.gfx_addr); |
1807 | POSTING_READ(RING_HWS_PGA(ring->mmio_base)); |
2437 | POSTING_READ(RING_HWS_PGA(ring->mmio_base)); |
1808 | } |
2438 | } |
1809 | 2439 | ||
1810 | /** |
2440 | /** |
1811 | * intel_lr_context_deferred_create() - create the LRC specific bits of a context |
2441 | * intel_lr_context_deferred_alloc() - create the LRC specific bits of a context |
1812 | * @ctx: LR context to create. |
2442 | * @ctx: LR context to create. |
1813 | * @ring: engine to be used with the context. |
2443 | * @ring: engine to be used with the context. |
1814 | * |
2444 | * |
1815 | * This function can be called more than once, with different engines, if we plan |
2445 | * This function can be called more than once, with different engines, if we plan |
1816 | * to use the context with them. The context backing objects and the ringbuffers |
2446 | * to use the context with them. The context backing objects and the ringbuffers |
1817 | * (specially the ringbuffer backing objects) suck a lot of memory up, and that's why |
2447 | * (specially the ringbuffer backing objects) suck a lot of memory up, and that's why |
1818 | * the creation is a deferred call: it's better to make sure first that we need to use |
2448 | * the creation is a deferred call: it's better to make sure first that we need to use |
1819 | * a given ring with the context. |
2449 | * a given ring with the context. |
1820 | * |
2450 | * |
1821 | * Return: non-zero on error. |
2451 | * Return: non-zero on error. |
1822 | */ |
2452 | */ |
- | 2453 | ||
1823 | int intel_lr_context_deferred_create(struct intel_context *ctx, |
2454 | int intel_lr_context_deferred_alloc(struct intel_context *ctx, |
1824 | struct intel_engine_cs *ring) |
2455 | struct intel_engine_cs *ring) |
1825 | { |
2456 | { |
1826 | const bool is_global_default_ctx = (ctx == ring->default_context); |
- | |
1827 | struct drm_device *dev = ring->dev; |
2457 | struct drm_device *dev = ring->dev; |
1828 | struct drm_i915_gem_object *ctx_obj; |
2458 | struct drm_i915_gem_object *ctx_obj; |
1829 | uint32_t context_size; |
2459 | uint32_t context_size; |
1830 | struct intel_ringbuffer *ringbuf; |
2460 | struct intel_ringbuffer *ringbuf; |
1831 | int ret; |
2461 | int ret; |
1832 | 2462 | ||
1833 | WARN_ON(ctx->legacy_hw_ctx.rcs_state != NULL); |
2463 | WARN_ON(ctx->legacy_hw_ctx.rcs_state != NULL); |
1834 | if (ctx->engine[ring->id].state) |
2464 | WARN_ON(ctx->engine[ring->id].state); |
1835 | return 0; |
- | |
1836 | 2465 | ||
1837 | context_size = round_up(get_lr_context_size(ring), 4096); |
2466 | context_size = round_up(get_lr_context_size(ring), 4096); |
1838 | 2467 | ||
1839 | ctx_obj = i915_gem_alloc_context_obj(dev, context_size); |
- | |
1840 | if (IS_ERR(ctx_obj)) { |
2468 | /* One extra page as the sharing data between driver and GuC */ |
1841 | ret = PTR_ERR(ctx_obj); |
- | |
1842 | DRM_DEBUG_DRIVER("Alloc LRC backing obj failed: %d\n", ret); |
- | |
1843 | return ret; |
- | |
1844 | } |
- | |
1845 | 2469 | context_size += PAGE_SIZE * LRC_PPHWSP_PN; |
|
1846 | if (is_global_default_ctx) { |
2470 | |
1847 | ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0); |
2471 | ctx_obj = i915_gem_alloc_object(dev, context_size); |
1848 | if (ret) { |
- | |
1849 | DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", |
- | |
1850 | ret); |
2472 | if (!ctx_obj) { |
1851 | drm_gem_object_unreference(&ctx_obj->base); |
2473 | DRM_DEBUG_DRIVER("Alloc LRC backing obj failed.\n"); |
- | 2474 | return -ENOMEM; |
|
- | 2475 | } |
|
- | 2476 | ||
- | 2477 | ringbuf = intel_engine_create_ringbuffer(ring, 4 * PAGE_SIZE); |
|
- | 2478 | if (IS_ERR(ringbuf)) { |
|
1852 | return ret; |
2479 | ret = PTR_ERR(ringbuf); |
1853 | } |
2480 | goto error_deref_obj; |
1854 | } |
2481 | } |
1855 | 2482 | ||
1856 | ringbuf = kzalloc(sizeof(*ringbuf), GFP_KERNEL); |
- | |
1857 | if (!ringbuf) { |
- | |
1858 | DRM_DEBUG_DRIVER("Failed to allocate ringbuffer %s\n", |
2483 | ret = populate_lr_context(ctx, ctx_obj, ring, ringbuf); |
1859 | ring->name); |
2484 | if (ret) { |
1860 | ret = -ENOMEM; |
2485 | DRM_DEBUG_DRIVER("Failed to populate LRC: %d\n", ret); |
1861 | goto error_unpin_ctx; |
2486 | goto error_ringbuf; |
1862 | } |
- | |
1863 | 2487 | } |
|
1864 | ringbuf->ring = ring; |
- | |
1865 | ringbuf->FIXME_lrc_ctx = ctx; |
- | |
1866 | - | ||
1867 | ringbuf->size = 32 * PAGE_SIZE; |
2488 | |
1868 | ringbuf->effective_size = ringbuf->size; |
2489 | ctx->engine[ring->id].ringbuf = ringbuf; |
1869 | ringbuf->head = 0; |
2490 | ctx->engine[ring->id].state = ctx_obj; |
1870 | ringbuf->tail = 0; |
2491 | |
1871 | ringbuf->space = ringbuf->size; |
- | |
1872 | ringbuf->last_retired_head = -1; |
2492 | if (ctx != ring->default_context && ring->init_context) { |
1873 | 2493 | struct drm_i915_gem_request *req; |
|
1874 | if (ringbuf->obj == NULL) { |
2494 | |
1875 | ret = intel_alloc_ringbuffer_obj(dev, ringbuf); |
2495 | ret = i915_gem_request_alloc(ring, |
1876 | if (ret) { |
2496 | ctx, &req); |
1877 | DRM_DEBUG_DRIVER( |
- | |
1878 | "Failed to allocate ringbuffer obj %s: %d\n", |
2497 | if (ret) { |
1879 | ring->name, ret); |
- | |
1880 | goto error_free_rbuf; |
2498 | DRM_ERROR("ring create req: %d\n", |
1881 | } |
2499 | ret); |
- | 2500 | goto error_ringbuf; |
|
1882 | 2501 | } |
|
1883 | if (is_global_default_ctx) { |
2502 | |
- | 2503 | ret = ring->init_context(req); |
|
1884 | ret = intel_pin_and_map_ringbuffer_obj(dev, ringbuf); |
2504 | if (ret) { |
- | 2505 | DRM_ERROR("ring init context: %d\n", |
|
- | 2506 | ret); |
|
- | 2507 | i915_gem_request_cancel(req); |
|
- | 2508 | goto error_ringbuf; |
|
- | 2509 | } |
|
- | 2510 | i915_add_request_no_flush(req); |
|
- | 2511 | } |
|
- | 2512 | return 0; |
|
1885 | if (ret) { |
2513 | |
1886 | DRM_ERROR( |
2514 | error_ringbuf: |
1887 | "Failed to pin and map ringbuffer %s: %d\n", |
2515 | intel_ringbuffer_free(ringbuf); |
- | 2516 | error_deref_obj: |
|
1888 | ring->name, ret); |
2517 | drm_gem_object_unreference(&ctx_obj->base); |
1889 | goto error_destroy_rbuf; |
2518 | ctx->engine[ring->id].ringbuf = NULL; |
1890 | } |
2519 | ctx->engine[ring->id].state = NULL; |
- | 2520 | return ret; |
|
- | 2521 | } |
|
1891 | } |
2522 | |
- | 2523 | void intel_lr_context_reset(struct drm_device *dev, |
|
1892 | 2524 | struct intel_context *ctx) |
|
- | 2525 | { |
|
- | 2526 | struct drm_i915_private *dev_priv = dev->dev_private; |
|
1893 | } |
2527 | struct intel_engine_cs *ring; |
1894 | 2528 | int i; |
|
1895 | ret = populate_lr_context(ctx, ctx_obj, ring, ringbuf); |
2529 | |
1896 | if (ret) { |
- | |
1897 | DRM_DEBUG_DRIVER("Failed to populate LRC: %d\n", ret); |
2530 | for_each_ring(ring, dev_priv, i) { |
1898 | goto error; |
2531 | struct drm_i915_gem_object *ctx_obj = |
1899 | } |
- | |
1900 | 2532 | ctx->engine[ring->id].state; |
|
- | 2533 | struct intel_ringbuffer *ringbuf = |
|
- | 2534 | ctx->engine[ring->id].ringbuf; |
|
1901 | ctx->engine[ring->id].ringbuf = ringbuf; |
- | |
1902 | ctx->engine[ring->id].state = ctx_obj; |
- | |
1903 | - | ||
1904 | if (ctx == ring->default_context) |
2535 | uint32_t *reg_state; |
1905 | lrc_setup_hardware_status_page(ring, ctx_obj); |
2536 | struct page *page; |
1906 | - | ||
1907 | if (ring->id == RCS && !ctx->rcs_initialized) { |
- | |
1908 | if (ring->init_context) { |
- | |
1909 | ret = ring->init_context(ring, ctx); |
- | |
1910 | if (ret) |
2537 | |
1911 | DRM_ERROR("ring init context: %d\n", ret); |
- | |
1912 | } |
- | |
1913 | - | ||
1914 | ret = intel_lr_context_render_state_init(ring, ctx); |
- | |
1915 | if (ret) { |
- | |
1916 | DRM_ERROR("Init render state failed: %d\n", ret); |
- | |
1917 | ctx->engine[ring->id].ringbuf = NULL; |
2538 | if (!ctx_obj) |
1918 | ctx->engine[ring->id].state = NULL; |
- | |
1919 | goto error; |
2539 | continue; |
1920 | } |
- | |
1921 | ctx->rcs_initialized = true; |
- | |
1922 | } |
2540 | |
1923 | 2541 | if (i915_gem_object_get_pages(ctx_obj)) { |
|
1924 | return 0; |
2542 | WARN(1, "Failed get_pages for context obj\n"); |
1925 | 2543 | continue; |
|
1926 | error: |
2544 | } |
1927 | if (is_global_default_ctx) |
2545 | page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); |
1928 | intel_unpin_ringbuffer_obj(ringbuf); |
2546 | reg_state = kmap_atomic(page); |
1929 | error_destroy_rbuf: |
2547 | |
1930 | intel_destroy_ringbuffer_obj(ringbuf); |
2548 | reg_state[CTX_RING_HEAD+1] = 0; |
1931 | error_free_rbuf: |
2549 | reg_state[CTX_RING_TAIL+1] = 0; |
1932 | kfree(ringbuf); |
2550 | |
1933 | error_unpin_ctx: |
2551 | kunmap_atomic(reg_state); |
1934 | if (is_global_default_ctx) |
2552 | |
1935 | i915_gem_object_ggtt_unpin(ctx_obj); |
2553 | ringbuf->head = 0; |
1936 | drm_gem_object_unreference(&ctx_obj->base); |
2554 | ringbuf->tail = 0; |
1937 | return ret; |
2555 | } |
1938 | }>>5); |
2556 | }>>><>5); |
1939 | ><5); |
2557 | ><5); |
1940 | >3)><3)>><>><>><>><>><>><>><>><>><>><>><>8)); |
2558 | > |
1941 | ><8)); |
2559 | >< |
1942 | >>>>><>><>>=>><>><>8) |
2560 | > |
1943 | enum><8) |
2561 | >< |
1944 | enum>5) |
2562 | > |
1945 | #define><5) |
2563 | >< |
1946 | #define>2) |
2564 | > |
1947 | #define><2) |
2565 | >< |
1948 | #define>1) |
2566 | >>>><>><>>><>><>><>><>>><>><>>><>><>><>8)><8)>=>=>>=>>=>>><>><>><>>=>><>><>=>8) |
1949 | #define><1) |
2567 | |
1950 | #define>0) |
2568 | #define><8) |
1951 | #define><0) |
2569 | |
1952 | #define>><>><>><>><>><>><>><>><>><>><>><>><> |
2570 | #define>5) |
- | 2571 | #define><5) |
|
- | 2572 | #define>2) |
|
- | 2573 | #define><2) |
|
- | 2574 | #define>1) |
|
- | 2575 | #define><1) |
|
- | 2576 | #define>0) |
|
- | 2577 | #define><0) |
|
- | 2578 | #define>><>><>><>><>><>><>><>><>><>><>><>><> |