Rev 6937 | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 6937 | Rev 7144 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright © 2011-2012 Intel Corporation |
2 | * Copyright © 2011-2012 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 | * |
25 | * |
26 | */ |
26 | */ |
27 | 27 | ||
28 | /* |
28 | /* |
29 | * This file implements HW context support. On gen5+ a HW context consists of an |
29 | * This file implements HW context support. On gen5+ a HW context consists of an |
30 | * opaque GPU object which is referenced at times of context saves and restores. |
30 | * opaque GPU object which is referenced at times of context saves and restores. |
31 | * With RC6 enabled, the context is also referenced as the GPU enters and exists |
31 | * With RC6 enabled, the context is also referenced as the GPU enters and exists |
32 | * from RC6 (GPU has it's own internal power context, except on gen5). Though |
32 | * from RC6 (GPU has it's own internal power context, except on gen5). Though |
33 | * something like a context does exist for the media ring, the code only |
33 | * something like a context does exist for the media ring, the code only |
34 | * supports contexts for the render ring. |
34 | * supports contexts for the render ring. |
35 | * |
35 | * |
36 | * In software, there is a distinction between contexts created by the user, |
36 | * In software, there is a distinction between contexts created by the user, |
37 | * and the default HW context. The default HW context is used by GPU clients |
37 | * and the default HW context. The default HW context is used by GPU clients |
38 | * that do not request setup of their own hardware context. The default |
38 | * that do not request setup of their own hardware context. The default |
39 | * context's state is never restored to help prevent programming errors. This |
39 | * context's state is never restored to help prevent programming errors. This |
40 | * would happen if a client ran and piggy-backed off another clients GPU state. |
40 | * would happen if a client ran and piggy-backed off another clients GPU state. |
41 | * The default context only exists to give the GPU some offset to load as the |
41 | * The default context only exists to give the GPU some offset to load as the |
42 | * current to invoke a save of the context we actually care about. In fact, the |
42 | * current to invoke a save of the context we actually care about. In fact, the |
43 | * code could likely be constructed, albeit in a more complicated fashion, to |
43 | * code could likely be constructed, albeit in a more complicated fashion, to |
44 | * never use the default context, though that limits the driver's ability to |
44 | * never use the default context, though that limits the driver's ability to |
45 | * swap out, and/or destroy other contexts. |
45 | * swap out, and/or destroy other contexts. |
46 | * |
46 | * |
47 | * All other contexts are created as a request by the GPU client. These contexts |
47 | * All other contexts are created as a request by the GPU client. These contexts |
48 | * store GPU state, and thus allow GPU clients to not re-emit state (and |
48 | * store GPU state, and thus allow GPU clients to not re-emit state (and |
49 | * potentially query certain state) at any time. The kernel driver makes |
49 | * potentially query certain state) at any time. The kernel driver makes |
50 | * certain that the appropriate commands are inserted. |
50 | * certain that the appropriate commands are inserted. |
51 | * |
51 | * |
52 | * The context life cycle is semi-complicated in that context BOs may live |
52 | * The context life cycle is semi-complicated in that context BOs may live |
53 | * longer than the context itself because of the way the hardware, and object |
53 | * longer than the context itself because of the way the hardware, and object |
54 | * tracking works. Below is a very crude representation of the state machine |
54 | * tracking works. Below is a very crude representation of the state machine |
55 | * describing the context life. |
55 | * describing the context life. |
56 | * refcount pincount active |
56 | * refcount pincount active |
57 | * S0: initial state 0 0 0 |
57 | * S0: initial state 0 0 0 |
58 | * S1: context created 1 0 0 |
58 | * S1: context created 1 0 0 |
59 | * S2: context is currently running 2 1 X |
59 | * S2: context is currently running 2 1 X |
60 | * S3: GPU referenced, but not current 2 0 1 |
60 | * S3: GPU referenced, but not current 2 0 1 |
61 | * S4: context is current, but destroyed 1 1 0 |
61 | * S4: context is current, but destroyed 1 1 0 |
62 | * S5: like S3, but destroyed 1 0 1 |
62 | * S5: like S3, but destroyed 1 0 1 |
63 | * |
63 | * |
64 | * The most common (but not all) transitions: |
64 | * The most common (but not all) transitions: |
65 | * S0->S1: client creates a context |
65 | * S0->S1: client creates a context |
66 | * S1->S2: client submits execbuf with context |
66 | * S1->S2: client submits execbuf with context |
67 | * S2->S3: other clients submits execbuf with context |
67 | * S2->S3: other clients submits execbuf with context |
68 | * S3->S1: context object was retired |
68 | * S3->S1: context object was retired |
69 | * S3->S2: clients submits another execbuf |
69 | * S3->S2: clients submits another execbuf |
70 | * S2->S4: context destroy called with current context |
70 | * S2->S4: context destroy called with current context |
71 | * S3->S5->S0: destroy path |
71 | * S3->S5->S0: destroy path |
72 | * S4->S5->S0: destroy path on current context |
72 | * S4->S5->S0: destroy path on current context |
73 | * |
73 | * |
74 | * There are two confusing terms used above: |
74 | * There are two confusing terms used above: |
75 | * The "current context" means the context which is currently running on the |
75 | * The "current context" means the context which is currently running on the |
76 | * GPU. The GPU has loaded its state already and has stored away the gtt |
76 | * GPU. The GPU has loaded its state already and has stored away the gtt |
77 | * offset of the BO. The GPU is not actively referencing the data at this |
77 | * offset of the BO. The GPU is not actively referencing the data at this |
78 | * offset, but it will on the next context switch. The only way to avoid this |
78 | * offset, but it will on the next context switch. The only way to avoid this |
79 | * is to do a GPU reset. |
79 | * is to do a GPU reset. |
80 | * |
80 | * |
81 | * An "active context' is one which was previously the "current context" and is |
81 | * An "active context' is one which was previously the "current context" and is |
82 | * on the active list waiting for the next context switch to occur. Until this |
82 | * on the active list waiting for the next context switch to occur. Until this |
83 | * happens, the object must remain at the same gtt offset. It is therefore |
83 | * happens, the object must remain at the same gtt offset. It is therefore |
84 | * possible to destroy a context, but it is still active. |
84 | * possible to destroy a context, but it is still active. |
85 | * |
85 | * |
86 | */ |
86 | */ |
87 | 87 | ||
88 | #include |
88 | #include |
89 | #include |
89 | #include |
90 | #include "i915_drv.h" |
90 | #include "i915_drv.h" |
91 | #include "i915_trace.h" |
91 | #include "i915_trace.h" |
92 | 92 | ||
93 | /* This is a HW constraint. The value below is the largest known requirement |
93 | /* This is a HW constraint. The value below is the largest known requirement |
94 | * I've seen in a spec to date, and that was a workaround for a non-shipping |
94 | * I've seen in a spec to date, and that was a workaround for a non-shipping |
95 | * part. It should be safe to decrease this, but it's more future proof as is. |
95 | * part. It should be safe to decrease this, but it's more future proof as is. |
96 | */ |
96 | */ |
97 | #define GEN6_CONTEXT_ALIGN (64<<10) |
97 | #define GEN6_CONTEXT_ALIGN (64<<10) |
98 | #define GEN7_CONTEXT_ALIGN 4096 |
98 | #define GEN7_CONTEXT_ALIGN 4096 |
99 | 99 | ||
100 | static size_t get_context_alignment(struct drm_device *dev) |
100 | static size_t get_context_alignment(struct drm_device *dev) |
101 | { |
101 | { |
102 | if (IS_GEN6(dev)) |
102 | if (IS_GEN6(dev)) |
103 | return GEN6_CONTEXT_ALIGN; |
103 | return GEN6_CONTEXT_ALIGN; |
104 | 104 | ||
105 | return GEN7_CONTEXT_ALIGN; |
105 | return GEN7_CONTEXT_ALIGN; |
106 | } |
106 | } |
107 | 107 | ||
108 | static int get_context_size(struct drm_device *dev) |
108 | static int get_context_size(struct drm_device *dev) |
109 | { |
109 | { |
110 | struct drm_i915_private *dev_priv = dev->dev_private; |
110 | struct drm_i915_private *dev_priv = dev->dev_private; |
111 | int ret; |
111 | int ret; |
112 | u32 reg; |
112 | u32 reg; |
113 | 113 | ||
114 | switch (INTEL_INFO(dev)->gen) { |
114 | switch (INTEL_INFO(dev)->gen) { |
115 | case 6: |
115 | case 6: |
116 | reg = I915_READ(CXT_SIZE); |
116 | reg = I915_READ(CXT_SIZE); |
117 | ret = GEN6_CXT_TOTAL_SIZE(reg) * 64; |
117 | ret = GEN6_CXT_TOTAL_SIZE(reg) * 64; |
118 | break; |
118 | break; |
119 | case 7: |
119 | case 7: |
120 | reg = I915_READ(GEN7_CXT_SIZE); |
120 | reg = I915_READ(GEN7_CXT_SIZE); |
121 | if (IS_HASWELL(dev)) |
121 | if (IS_HASWELL(dev)) |
122 | ret = HSW_CXT_TOTAL_SIZE; |
122 | ret = HSW_CXT_TOTAL_SIZE; |
123 | else |
123 | else |
124 | ret = GEN7_CXT_TOTAL_SIZE(reg) * 64; |
124 | ret = GEN7_CXT_TOTAL_SIZE(reg) * 64; |
125 | break; |
125 | break; |
126 | case 8: |
126 | case 8: |
127 | ret = GEN8_CXT_TOTAL_SIZE; |
127 | ret = GEN8_CXT_TOTAL_SIZE; |
128 | break; |
128 | break; |
129 | default: |
129 | default: |
130 | BUG(); |
130 | BUG(); |
131 | } |
131 | } |
132 | 132 | ||
133 | return ret; |
133 | return ret; |
134 | } |
134 | } |
135 | 135 | ||
136 | static void i915_gem_context_clean(struct intel_context *ctx) |
136 | static void i915_gem_context_clean(struct intel_context *ctx) |
137 | { |
137 | { |
138 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; |
138 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; |
139 | struct i915_vma *vma, *next; |
139 | struct i915_vma *vma, *next; |
140 | 140 | ||
141 | if (!ppgtt) |
141 | if (!ppgtt) |
142 | return; |
142 | return; |
143 | 143 | ||
144 | list_for_each_entry_safe(vma, next, &ppgtt->base.inactive_list, |
144 | list_for_each_entry_safe(vma, next, &ppgtt->base.inactive_list, |
145 | mm_list) { |
145 | vm_link) { |
146 | if (WARN_ON(__i915_vma_unbind_no_wait(vma))) |
146 | if (WARN_ON(__i915_vma_unbind_no_wait(vma))) |
147 | break; |
147 | break; |
148 | } |
148 | } |
149 | } |
149 | } |
150 | 150 | ||
151 | void i915_gem_context_free(struct kref *ctx_ref) |
151 | void i915_gem_context_free(struct kref *ctx_ref) |
152 | { |
152 | { |
153 | struct intel_context *ctx = container_of(ctx_ref, typeof(*ctx), ref); |
153 | struct intel_context *ctx = container_of(ctx_ref, typeof(*ctx), ref); |
154 | 154 | ||
155 | trace_i915_context_free(ctx); |
155 | trace_i915_context_free(ctx); |
156 | 156 | ||
157 | if (i915.enable_execlists) |
157 | if (i915.enable_execlists) |
158 | intel_lr_context_free(ctx); |
158 | intel_lr_context_free(ctx); |
159 | 159 | ||
160 | /* |
160 | /* |
161 | * This context is going away and we need to remove all VMAs still |
161 | * This context is going away and we need to remove all VMAs still |
162 | * around. This is to handle imported shared objects for which |
162 | * around. This is to handle imported shared objects for which |
163 | * destructor did not run when their handles were closed. |
163 | * destructor did not run when their handles were closed. |
164 | */ |
164 | */ |
165 | i915_gem_context_clean(ctx); |
165 | i915_gem_context_clean(ctx); |
166 | 166 | ||
167 | i915_ppgtt_put(ctx->ppgtt); |
167 | i915_ppgtt_put(ctx->ppgtt); |
168 | 168 | ||
169 | if (ctx->legacy_hw_ctx.rcs_state) |
169 | if (ctx->legacy_hw_ctx.rcs_state) |
170 | drm_gem_object_unreference(&ctx->legacy_hw_ctx.rcs_state->base); |
170 | drm_gem_object_unreference(&ctx->legacy_hw_ctx.rcs_state->base); |
171 | list_del(&ctx->link); |
171 | list_del(&ctx->link); |
172 | kfree(ctx); |
172 | kfree(ctx); |
173 | } |
173 | } |
174 | 174 | ||
175 | struct drm_i915_gem_object * |
175 | struct drm_i915_gem_object * |
176 | i915_gem_alloc_context_obj(struct drm_device *dev, size_t size) |
176 | i915_gem_alloc_context_obj(struct drm_device *dev, size_t size) |
177 | { |
177 | { |
178 | struct drm_i915_gem_object *obj; |
178 | struct drm_i915_gem_object *obj; |
179 | int ret; |
179 | int ret; |
180 | 180 | ||
181 | obj = i915_gem_alloc_object(dev, size); |
181 | obj = i915_gem_alloc_object(dev, size); |
182 | if (obj == NULL) |
182 | if (obj == NULL) |
183 | return ERR_PTR(-ENOMEM); |
183 | return ERR_PTR(-ENOMEM); |
184 | 184 | ||
185 | /* |
185 | /* |
186 | * Try to make the context utilize L3 as well as LLC. |
186 | * Try to make the context utilize L3 as well as LLC. |
187 | * |
187 | * |
188 | * On VLV we don't have L3 controls in the PTEs so we |
188 | * On VLV we don't have L3 controls in the PTEs so we |
189 | * shouldn't touch the cache level, especially as that |
189 | * shouldn't touch the cache level, especially as that |
190 | * would make the object snooped which might have a |
190 | * would make the object snooped which might have a |
191 | * negative performance impact. |
191 | * negative performance impact. |
192 | * |
192 | * |
193 | * Snooping is required on non-llc platforms in execlist |
193 | * Snooping is required on non-llc platforms in execlist |
194 | * mode, but since all GGTT accesses use PAT entry 0 we |
194 | * mode, but since all GGTT accesses use PAT entry 0 we |
195 | * get snooping anyway regardless of cache_level. |
195 | * get snooping anyway regardless of cache_level. |
196 | * |
196 | * |
197 | * This is only applicable for Ivy Bridge devices since |
197 | * This is only applicable for Ivy Bridge devices since |
198 | * later platforms don't have L3 control bits in the PTE. |
198 | * later platforms don't have L3 control bits in the PTE. |
199 | */ |
199 | */ |
200 | if (IS_IVYBRIDGE(dev)) { |
200 | if (IS_IVYBRIDGE(dev)) { |
201 | ret = i915_gem_object_set_cache_level(obj, I915_CACHE_L3_LLC); |
201 | ret = i915_gem_object_set_cache_level(obj, I915_CACHE_L3_LLC); |
202 | /* Failure shouldn't ever happen this early */ |
202 | /* Failure shouldn't ever happen this early */ |
203 | if (WARN_ON(ret)) { |
203 | if (WARN_ON(ret)) { |
204 | drm_gem_object_unreference(&obj->base); |
204 | drm_gem_object_unreference(&obj->base); |
205 | return ERR_PTR(ret); |
205 | return ERR_PTR(ret); |
206 | } |
206 | } |
207 | } |
207 | } |
208 | 208 | ||
209 | return obj; |
209 | return obj; |
210 | } |
210 | } |
211 | 211 | ||
212 | static struct intel_context * |
212 | static struct intel_context * |
213 | __create_hw_context(struct drm_device *dev, |
213 | __create_hw_context(struct drm_device *dev, |
214 | struct drm_i915_file_private *file_priv) |
214 | struct drm_i915_file_private *file_priv) |
215 | { |
215 | { |
216 | struct drm_i915_private *dev_priv = dev->dev_private; |
216 | struct drm_i915_private *dev_priv = dev->dev_private; |
217 | struct intel_context *ctx; |
217 | struct intel_context *ctx; |
218 | int ret; |
218 | int ret; |
219 | 219 | ||
220 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); |
220 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); |
221 | if (ctx == NULL) |
221 | if (ctx == NULL) |
222 | return ERR_PTR(-ENOMEM); |
222 | return ERR_PTR(-ENOMEM); |
223 | 223 | ||
224 | kref_init(&ctx->ref); |
224 | kref_init(&ctx->ref); |
225 | list_add_tail(&ctx->link, &dev_priv->context_list); |
225 | list_add_tail(&ctx->link, &dev_priv->context_list); |
226 | ctx->i915 = dev_priv; |
226 | ctx->i915 = dev_priv; |
227 | 227 | ||
228 | if (dev_priv->hw_context_size) { |
228 | if (dev_priv->hw_context_size) { |
229 | struct drm_i915_gem_object *obj = |
229 | struct drm_i915_gem_object *obj = |
230 | i915_gem_alloc_context_obj(dev, dev_priv->hw_context_size); |
230 | i915_gem_alloc_context_obj(dev, dev_priv->hw_context_size); |
231 | if (IS_ERR(obj)) { |
231 | if (IS_ERR(obj)) { |
232 | ret = PTR_ERR(obj); |
232 | ret = PTR_ERR(obj); |
233 | goto err_out; |
233 | goto err_out; |
234 | } |
234 | } |
235 | ctx->legacy_hw_ctx.rcs_state = obj; |
235 | ctx->legacy_hw_ctx.rcs_state = obj; |
236 | } |
236 | } |
237 | 237 | ||
238 | /* Default context will never have a file_priv */ |
238 | /* Default context will never have a file_priv */ |
239 | if (file_priv != NULL) { |
239 | if (file_priv != NULL) { |
240 | ret = idr_alloc(&file_priv->context_idr, ctx, |
240 | ret = idr_alloc(&file_priv->context_idr, ctx, |
241 | DEFAULT_CONTEXT_HANDLE, 0, GFP_KERNEL); |
241 | DEFAULT_CONTEXT_HANDLE, 0, GFP_KERNEL); |
242 | if (ret < 0) |
242 | if (ret < 0) |
243 | goto err_out; |
243 | goto err_out; |
244 | } else |
244 | } else |
245 | ret = DEFAULT_CONTEXT_HANDLE; |
245 | ret = DEFAULT_CONTEXT_HANDLE; |
246 | 246 | ||
247 | ctx->file_priv = file_priv; |
247 | ctx->file_priv = file_priv; |
248 | ctx->user_handle = ret; |
248 | ctx->user_handle = ret; |
249 | /* NB: Mark all slices as needing a remap so that when the context first |
249 | /* NB: Mark all slices as needing a remap so that when the context first |
250 | * loads it will restore whatever remap state already exists. If there |
250 | * loads it will restore whatever remap state already exists. If there |
251 | * is no remap info, it will be a NOP. */ |
251 | * is no remap info, it will be a NOP. */ |
252 | ctx->remap_slice = (1 << NUM_L3_SLICES(dev)) - 1; |
252 | ctx->remap_slice = (1 << NUM_L3_SLICES(dev)) - 1; |
253 | 253 | ||
254 | ctx->hang_stats.ban_period_seconds = DRM_I915_CTX_BAN_PERIOD; |
254 | ctx->hang_stats.ban_period_seconds = DRM_I915_CTX_BAN_PERIOD; |
255 | 255 | ||
256 | return ctx; |
256 | return ctx; |
257 | 257 | ||
258 | err_out: |
258 | err_out: |
259 | i915_gem_context_unreference(ctx); |
259 | i915_gem_context_unreference(ctx); |
260 | return ERR_PTR(ret); |
260 | return ERR_PTR(ret); |
261 | } |
261 | } |
262 | 262 | ||
263 | /** |
263 | /** |
264 | * The default context needs to exist per ring that uses contexts. It stores the |
264 | * The default context needs to exist per ring that uses contexts. It stores the |
265 | * context state of the GPU for applications that don't utilize HW contexts, as |
265 | * context state of the GPU for applications that don't utilize HW contexts, as |
266 | * well as an idle case. |
266 | * well as an idle case. |
267 | */ |
267 | */ |
268 | static struct intel_context * |
268 | static struct intel_context * |
269 | i915_gem_create_context(struct drm_device *dev, |
269 | i915_gem_create_context(struct drm_device *dev, |
270 | struct drm_i915_file_private *file_priv) |
270 | struct drm_i915_file_private *file_priv) |
271 | { |
271 | { |
272 | const bool is_global_default_ctx = file_priv == NULL; |
272 | const bool is_global_default_ctx = file_priv == NULL; |
273 | struct intel_context *ctx; |
273 | struct intel_context *ctx; |
274 | int ret = 0; |
274 | int ret = 0; |
275 | 275 | ||
276 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
276 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
277 | 277 | ||
278 | ctx = __create_hw_context(dev, file_priv); |
278 | ctx = __create_hw_context(dev, file_priv); |
279 | if (IS_ERR(ctx)) |
279 | if (IS_ERR(ctx)) |
280 | return ctx; |
280 | return ctx; |
281 | 281 | ||
282 | if (is_global_default_ctx && ctx->legacy_hw_ctx.rcs_state) { |
282 | if (is_global_default_ctx && ctx->legacy_hw_ctx.rcs_state) { |
283 | /* We may need to do things with the shrinker which |
283 | /* We may need to do things with the shrinker which |
284 | * require us to immediately switch back to the default |
284 | * require us to immediately switch back to the default |
285 | * context. This can cause a problem as pinning the |
285 | * context. This can cause a problem as pinning the |
286 | * default context also requires GTT space which may not |
286 | * default context also requires GTT space which may not |
287 | * be available. To avoid this we always pin the default |
287 | * be available. To avoid this we always pin the default |
288 | * context. |
288 | * context. |
289 | */ |
289 | */ |
290 | ret = i915_gem_obj_ggtt_pin(ctx->legacy_hw_ctx.rcs_state, |
290 | ret = i915_gem_obj_ggtt_pin(ctx->legacy_hw_ctx.rcs_state, |
291 | get_context_alignment(dev), 0); |
291 | get_context_alignment(dev), 0); |
292 | if (ret) { |
292 | if (ret) { |
293 | DRM_DEBUG_DRIVER("Couldn't pin %d\n", ret); |
293 | DRM_DEBUG_DRIVER("Couldn't pin %d\n", ret); |
294 | goto err_destroy; |
294 | goto err_destroy; |
295 | } |
295 | } |
296 | } |
296 | } |
297 | 297 | ||
298 | if (USES_FULL_PPGTT(dev)) { |
298 | if (USES_FULL_PPGTT(dev)) { |
299 | struct i915_hw_ppgtt *ppgtt = i915_ppgtt_create(dev, file_priv); |
299 | struct i915_hw_ppgtt *ppgtt = i915_ppgtt_create(dev, file_priv); |
300 | 300 | ||
301 | if (IS_ERR_OR_NULL(ppgtt)) { |
301 | if (IS_ERR_OR_NULL(ppgtt)) { |
302 | DRM_DEBUG_DRIVER("PPGTT setup failed (%ld)\n", |
302 | DRM_DEBUG_DRIVER("PPGTT setup failed (%ld)\n", |
303 | PTR_ERR(ppgtt)); |
303 | PTR_ERR(ppgtt)); |
304 | ret = PTR_ERR(ppgtt); |
304 | ret = PTR_ERR(ppgtt); |
305 | goto err_unpin; |
305 | goto err_unpin; |
306 | } |
306 | } |
307 | 307 | ||
308 | ctx->ppgtt = ppgtt; |
308 | ctx->ppgtt = ppgtt; |
309 | } |
309 | } |
310 | 310 | ||
311 | trace_i915_context_create(ctx); |
311 | trace_i915_context_create(ctx); |
312 | 312 | ||
313 | return ctx; |
313 | return ctx; |
314 | 314 | ||
315 | err_unpin: |
315 | err_unpin: |
316 | if (is_global_default_ctx && ctx->legacy_hw_ctx.rcs_state) |
316 | if (is_global_default_ctx && ctx->legacy_hw_ctx.rcs_state) |
317 | i915_gem_object_ggtt_unpin(ctx->legacy_hw_ctx.rcs_state); |
317 | i915_gem_object_ggtt_unpin(ctx->legacy_hw_ctx.rcs_state); |
318 | err_destroy: |
318 | err_destroy: |
319 | idr_remove(&file_priv->context_idr, ctx->user_handle); |
319 | idr_remove(&file_priv->context_idr, ctx->user_handle); |
320 | i915_gem_context_unreference(ctx); |
320 | i915_gem_context_unreference(ctx); |
321 | return ERR_PTR(ret); |
321 | return ERR_PTR(ret); |
322 | } |
322 | } |
- | 323 | ||
- | 324 | static void i915_gem_context_unpin(struct intel_context *ctx, |
|
- | 325 | struct intel_engine_cs *engine) |
|
- | 326 | { |
|
- | 327 | if (i915.enable_execlists) { |
|
- | 328 | intel_lr_context_unpin(ctx, engine); |
|
- | 329 | } else { |
|
- | 330 | if (engine->id == RCS && ctx->legacy_hw_ctx.rcs_state) |
|
- | 331 | i915_gem_object_ggtt_unpin(ctx->legacy_hw_ctx.rcs_state); |
|
- | 332 | i915_gem_context_unreference(ctx); |
|
- | 333 | } |
|
- | 334 | } |
|
323 | 335 | ||
324 | void i915_gem_context_reset(struct drm_device *dev) |
336 | void i915_gem_context_reset(struct drm_device *dev) |
325 | { |
337 | { |
326 | struct drm_i915_private *dev_priv = dev->dev_private; |
338 | struct drm_i915_private *dev_priv = dev->dev_private; |
327 | int i; |
339 | int i; |
328 | 340 | ||
329 | if (i915.enable_execlists) { |
341 | if (i915.enable_execlists) { |
330 | struct intel_context *ctx; |
342 | struct intel_context *ctx; |
331 | 343 | ||
332 | list_for_each_entry(ctx, &dev_priv->context_list, link) { |
344 | list_for_each_entry(ctx, &dev_priv->context_list, link) |
333 | intel_lr_context_reset(dev, ctx); |
345 | intel_lr_context_reset(dev, ctx); |
334 | } |
346 | } |
335 | - | ||
336 | return; |
- | |
337 | } |
- | |
338 | 347 | ||
339 | for (i = 0; i < I915_NUM_RINGS; i++) { |
348 | for (i = 0; i < I915_NUM_RINGS; i++) { |
340 | struct intel_engine_cs *ring = &dev_priv->ring[i]; |
- | |
341 | struct intel_context *lctx = ring->last_context; |
- | |
342 | - | ||
343 | if (lctx) { |
- | |
344 | if (lctx->legacy_hw_ctx.rcs_state && i == RCS) |
- | |
- | 349 | struct intel_engine_cs *ring = &dev_priv->ring[i]; |
|
345 | i915_gem_object_ggtt_unpin(lctx->legacy_hw_ctx.rcs_state); |
350 | |
346 | 351 | if (ring->last_context) { |
|
347 | i915_gem_context_unreference(lctx); |
352 | i915_gem_context_unpin(ring->last_context, ring); |
- | 353 | ring->last_context = NULL; |
|
348 | ring->last_context = NULL; |
354 | } |
349 | } |
- | |
350 | 355 | } |
|
351 | /* Force the GPU state to be reinitialised on enabling */ |
- | |
352 | if (ring->default_context) |
356 | |
353 | ring->default_context->legacy_hw_ctx.initialized = false; |
357 | /* Force the GPU state to be reinitialised on enabling */ |
354 | } |
358 | dev_priv->kernel_context->legacy_hw_ctx.initialized = false; |
355 | } |
359 | } |
356 | 360 | ||
357 | int i915_gem_context_init(struct drm_device *dev) |
361 | int i915_gem_context_init(struct drm_device *dev) |
358 | { |
362 | { |
359 | struct drm_i915_private *dev_priv = dev->dev_private; |
363 | struct drm_i915_private *dev_priv = dev->dev_private; |
360 | struct intel_context *ctx; |
364 | struct intel_context *ctx; |
361 | int i; |
- | |
362 | 365 | ||
363 | /* Init should only be called once per module load. Eventually the |
366 | /* Init should only be called once per module load. Eventually the |
364 | * restriction on the context_disabled check can be loosened. */ |
367 | * restriction on the context_disabled check can be loosened. */ |
365 | if (WARN_ON(dev_priv->ring[RCS].default_context)) |
368 | if (WARN_ON(dev_priv->kernel_context)) |
366 | return 0; |
369 | return 0; |
367 | 370 | ||
368 | if (intel_vgpu_active(dev) && HAS_LOGICAL_RING_CONTEXTS(dev)) { |
371 | if (intel_vgpu_active(dev) && HAS_LOGICAL_RING_CONTEXTS(dev)) { |
369 | if (!i915.enable_execlists) { |
372 | if (!i915.enable_execlists) { |
370 | DRM_INFO("Only EXECLIST mode is supported in vgpu.\n"); |
373 | DRM_INFO("Only EXECLIST mode is supported in vgpu.\n"); |
371 | return -EINVAL; |
374 | return -EINVAL; |
372 | } |
375 | } |
373 | } |
376 | } |
374 | 377 | ||
375 | if (i915.enable_execlists) { |
378 | if (i915.enable_execlists) { |
376 | /* NB: intentionally left blank. We will allocate our own |
379 | /* NB: intentionally left blank. We will allocate our own |
377 | * backing objects as we need them, thank you very much */ |
380 | * backing objects as we need them, thank you very much */ |
378 | dev_priv->hw_context_size = 0; |
381 | dev_priv->hw_context_size = 0; |
379 | } else if (HAS_HW_CONTEXTS(dev)) { |
382 | } else if (HAS_HW_CONTEXTS(dev)) { |
380 | dev_priv->hw_context_size = round_up(get_context_size(dev), 4096); |
383 | dev_priv->hw_context_size = round_up(get_context_size(dev), 4096); |
381 | if (dev_priv->hw_context_size > (1<<20)) { |
384 | if (dev_priv->hw_context_size > (1<<20)) { |
382 | DRM_DEBUG_DRIVER("Disabling HW Contexts; invalid size %d\n", |
385 | DRM_DEBUG_DRIVER("Disabling HW Contexts; invalid size %d\n", |
383 | dev_priv->hw_context_size); |
386 | dev_priv->hw_context_size); |
384 | dev_priv->hw_context_size = 0; |
387 | dev_priv->hw_context_size = 0; |
385 | } |
388 | } |
386 | } |
389 | } |
387 | 390 | ||
388 | ctx = i915_gem_create_context(dev, NULL); |
391 | ctx = i915_gem_create_context(dev, NULL); |
389 | if (IS_ERR(ctx)) { |
392 | if (IS_ERR(ctx)) { |
390 | DRM_ERROR("Failed to create default global context (error %ld)\n", |
393 | DRM_ERROR("Failed to create default global context (error %ld)\n", |
391 | PTR_ERR(ctx)); |
394 | PTR_ERR(ctx)); |
392 | return PTR_ERR(ctx); |
395 | return PTR_ERR(ctx); |
393 | } |
396 | } |
394 | - | ||
395 | for (i = 0; i < I915_NUM_RINGS; i++) { |
- | |
396 | struct intel_engine_cs *ring = &dev_priv->ring[i]; |
- | |
397 | - | ||
398 | /* NB: RCS will hold a ref for all rings */ |
397 | |
399 | ring->default_context = ctx; |
- | |
400 | } |
398 | dev_priv->kernel_context = ctx; |
401 | 399 | ||
402 | DRM_DEBUG_DRIVER("%s context support initialized\n", |
400 | DRM_DEBUG_DRIVER("%s context support initialized\n", |
403 | i915.enable_execlists ? "LR" : |
401 | i915.enable_execlists ? "LR" : |
404 | dev_priv->hw_context_size ? "HW" : "fake"); |
402 | dev_priv->hw_context_size ? "HW" : "fake"); |
405 | return 0; |
403 | return 0; |
406 | } |
404 | } |
407 | 405 | ||
408 | void i915_gem_context_fini(struct drm_device *dev) |
406 | void i915_gem_context_fini(struct drm_device *dev) |
409 | { |
407 | { |
410 | struct drm_i915_private *dev_priv = dev->dev_private; |
408 | struct drm_i915_private *dev_priv = dev->dev_private; |
411 | struct intel_context *dctx = dev_priv->ring[RCS].default_context; |
409 | struct intel_context *dctx = dev_priv->kernel_context; |
412 | int i; |
410 | int i; |
413 | 411 | ||
414 | if (dctx->legacy_hw_ctx.rcs_state) { |
412 | if (dctx->legacy_hw_ctx.rcs_state) { |
415 | /* The only known way to stop the gpu from accessing the hw context is |
413 | /* The only known way to stop the gpu from accessing the hw context is |
416 | * to reset it. Do this as the very last operation to avoid confusing |
414 | * to reset it. Do this as the very last operation to avoid confusing |
417 | * other code, leading to spurious errors. */ |
415 | * other code, leading to spurious errors. */ |
418 | intel_gpu_reset(dev); |
416 | intel_gpu_reset(dev); |
419 | 417 | ||
420 | /* When default context is created and switched to, base object refcount |
418 | /* When default context is created and switched to, base object refcount |
421 | * will be 2 (+1 from object creation and +1 from do_switch()). |
419 | * will be 2 (+1 from object creation and +1 from do_switch()). |
422 | * i915_gem_context_fini() will be called after gpu_idle() has switched |
420 | * i915_gem_context_fini() will be called after gpu_idle() has switched |
423 | * to default context. So we need to unreference the base object once |
421 | * to default context. So we need to unreference the base object once |
424 | * to offset the do_switch part, so that i915_gem_context_unreference() |
422 | * to offset the do_switch part, so that i915_gem_context_unreference() |
425 | * can then free the base object correctly. */ |
423 | * can then free the base object correctly. */ |
426 | WARN_ON(!dev_priv->ring[RCS].last_context); |
424 | WARN_ON(!dev_priv->ring[RCS].last_context); |
427 | if (dev_priv->ring[RCS].last_context == dctx) { |
- | |
428 | /* Fake switch to NULL context */ |
- | |
429 | WARN_ON(dctx->legacy_hw_ctx.rcs_state->active); |
- | |
430 | i915_gem_object_ggtt_unpin(dctx->legacy_hw_ctx.rcs_state); |
- | |
431 | i915_gem_context_unreference(dctx); |
- | |
432 | dev_priv->ring[RCS].last_context = NULL; |
- | |
433 | } |
- | |
434 | 425 | ||
435 | i915_gem_object_ggtt_unpin(dctx->legacy_hw_ctx.rcs_state); |
426 | i915_gem_object_ggtt_unpin(dctx->legacy_hw_ctx.rcs_state); |
436 | } |
427 | } |
437 | 428 | ||
438 | for (i = 0; i < I915_NUM_RINGS; i++) { |
429 | for (i = I915_NUM_RINGS; --i >= 0;) { |
439 | struct intel_engine_cs *ring = &dev_priv->ring[i]; |
430 | struct intel_engine_cs *ring = &dev_priv->ring[i]; |
440 | 431 | ||
441 | if (ring->last_context) |
- | |
442 | i915_gem_context_unreference(ring->last_context); |
- | |
443 | 432 | if (ring->last_context) { |
|
444 | ring->default_context = NULL; |
433 | i915_gem_context_unpin(ring->last_context, ring); |
- | 434 | ring->last_context = NULL; |
|
445 | ring->last_context = NULL; |
435 | } |
- | 436 | } |
|
446 | } |
437 | |
447 | 438 | i915_gem_context_unreference(dctx); |
|
448 | i915_gem_context_unreference(dctx); |
439 | dev_priv->kernel_context = NULL; |
449 | } |
440 | } |
450 | 441 | ||
451 | int i915_gem_context_enable(struct drm_i915_gem_request *req) |
442 | int i915_gem_context_enable(struct drm_i915_gem_request *req) |
452 | { |
443 | { |
453 | struct intel_engine_cs *ring = req->ring; |
444 | struct intel_engine_cs *ring = req->ring; |
454 | int ret; |
445 | int ret; |
455 | 446 | ||
456 | if (i915.enable_execlists) { |
447 | if (i915.enable_execlists) { |
457 | if (ring->init_context == NULL) |
448 | if (ring->init_context == NULL) |
458 | return 0; |
449 | return 0; |
459 | 450 | ||
460 | ret = ring->init_context(req); |
451 | ret = ring->init_context(req); |
461 | } else |
452 | } else |
462 | ret = i915_switch_context(req); |
453 | ret = i915_switch_context(req); |
463 | 454 | ||
464 | if (ret) { |
455 | if (ret) { |
465 | DRM_ERROR("ring init context: %d\n", ret); |
456 | DRM_ERROR("ring init context: %d\n", ret); |
466 | return ret; |
457 | return ret; |
467 | } |
458 | } |
468 | 459 | ||
469 | return 0; |
460 | return 0; |
470 | } |
461 | } |
471 | 462 | ||
472 | static int context_idr_cleanup(int id, void *p, void *data) |
463 | static int context_idr_cleanup(int id, void *p, void *data) |
473 | { |
464 | { |
474 | struct intel_context *ctx = p; |
465 | struct intel_context *ctx = p; |
475 | 466 | ||
476 | i915_gem_context_unreference(ctx); |
467 | i915_gem_context_unreference(ctx); |
477 | return 0; |
468 | return 0; |
478 | } |
469 | } |
479 | 470 | ||
480 | int i915_gem_context_open(struct drm_device *dev, struct drm_file *file) |
471 | int i915_gem_context_open(struct drm_device *dev, struct drm_file *file) |
481 | { |
472 | { |
482 | struct drm_i915_file_private *file_priv = file->driver_priv; |
473 | struct drm_i915_file_private *file_priv = file->driver_priv; |
483 | struct intel_context *ctx; |
474 | struct intel_context *ctx; |
484 | 475 | ||
485 | idr_init(&file_priv->context_idr); |
476 | idr_init(&file_priv->context_idr); |
486 | 477 | ||
487 | mutex_lock(&dev->struct_mutex); |
478 | mutex_lock(&dev->struct_mutex); |
488 | ctx = i915_gem_create_context(dev, file_priv); |
479 | ctx = i915_gem_create_context(dev, file_priv); |
489 | mutex_unlock(&dev->struct_mutex); |
480 | mutex_unlock(&dev->struct_mutex); |
490 | 481 | ||
491 | if (IS_ERR(ctx)) { |
482 | if (IS_ERR(ctx)) { |
492 | idr_destroy(&file_priv->context_idr); |
483 | idr_destroy(&file_priv->context_idr); |
493 | return PTR_ERR(ctx); |
484 | return PTR_ERR(ctx); |
494 | } |
485 | } |
495 | 486 | ||
496 | return 0; |
487 | return 0; |
497 | } |
488 | } |
498 | 489 | ||
499 | void i915_gem_context_close(struct drm_device *dev, struct drm_file *file) |
490 | void i915_gem_context_close(struct drm_device *dev, struct drm_file *file) |
500 | { |
491 | { |
501 | struct drm_i915_file_private *file_priv = file->driver_priv; |
492 | struct drm_i915_file_private *file_priv = file->driver_priv; |
502 | 493 | ||
503 | idr_for_each(&file_priv->context_idr, context_idr_cleanup, NULL); |
494 | idr_for_each(&file_priv->context_idr, context_idr_cleanup, NULL); |
504 | idr_destroy(&file_priv->context_idr); |
495 | idr_destroy(&file_priv->context_idr); |
505 | } |
496 | } |
506 | 497 | ||
507 | struct intel_context * |
498 | struct intel_context * |
508 | i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id) |
499 | i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id) |
509 | { |
500 | { |
510 | struct intel_context *ctx; |
501 | struct intel_context *ctx; |
511 | 502 | ||
512 | ctx = (struct intel_context *)idr_find(&file_priv->context_idr, id); |
503 | ctx = (struct intel_context *)idr_find(&file_priv->context_idr, id); |
513 | if (!ctx) |
504 | if (!ctx) |
514 | return ERR_PTR(-ENOENT); |
505 | return ERR_PTR(-ENOENT); |
515 | 506 | ||
516 | return ctx; |
507 | return ctx; |
517 | } |
508 | } |
518 | 509 | ||
519 | static inline int |
510 | static inline int |
520 | mi_set_context(struct drm_i915_gem_request *req, u32 hw_flags) |
511 | mi_set_context(struct drm_i915_gem_request *req, u32 hw_flags) |
521 | { |
512 | { |
522 | struct intel_engine_cs *ring = req->ring; |
513 | struct intel_engine_cs *ring = req->ring; |
523 | u32 flags = hw_flags | MI_MM_SPACE_GTT; |
514 | u32 flags = hw_flags | MI_MM_SPACE_GTT; |
524 | const int num_rings = |
515 | const int num_rings = |
525 | /* Use an extended w/a on ivb+ if signalling from other rings */ |
516 | /* Use an extended w/a on ivb+ if signalling from other rings */ |
526 | i915_semaphore_is_enabled(ring->dev) ? |
517 | i915_semaphore_is_enabled(ring->dev) ? |
527 | hweight32(INTEL_INFO(ring->dev)->ring_mask) - 1 : |
518 | hweight32(INTEL_INFO(ring->dev)->ring_mask) - 1 : |
528 | 0; |
519 | 0; |
529 | int len, i, ret; |
520 | int len, i, ret; |
530 | 521 | ||
531 | /* w/a: If Flush TLB Invalidation Mode is enabled, driver must do a TLB |
522 | /* w/a: If Flush TLB Invalidation Mode is enabled, driver must do a TLB |
532 | * invalidation prior to MI_SET_CONTEXT. On GEN6 we don't set the value |
523 | * invalidation prior to MI_SET_CONTEXT. On GEN6 we don't set the value |
533 | * explicitly, so we rely on the value at ring init, stored in |
524 | * explicitly, so we rely on the value at ring init, stored in |
534 | * itlb_before_ctx_switch. |
525 | * itlb_before_ctx_switch. |
535 | */ |
526 | */ |
536 | if (IS_GEN6(ring->dev)) { |
527 | if (IS_GEN6(ring->dev)) { |
537 | ret = ring->flush(req, I915_GEM_GPU_DOMAINS, 0); |
528 | ret = ring->flush(req, I915_GEM_GPU_DOMAINS, 0); |
538 | if (ret) |
529 | if (ret) |
539 | return ret; |
530 | return ret; |
540 | } |
531 | } |
541 | 532 | ||
542 | /* These flags are for resource streamer on HSW+ */ |
533 | /* These flags are for resource streamer on HSW+ */ |
543 | if (IS_HASWELL(ring->dev) || INTEL_INFO(ring->dev)->gen >= 8) |
534 | if (IS_HASWELL(ring->dev) || INTEL_INFO(ring->dev)->gen >= 8) |
544 | flags |= (HSW_MI_RS_SAVE_STATE_EN | HSW_MI_RS_RESTORE_STATE_EN); |
535 | flags |= (HSW_MI_RS_SAVE_STATE_EN | HSW_MI_RS_RESTORE_STATE_EN); |
545 | else if (INTEL_INFO(ring->dev)->gen < 8) |
536 | else if (INTEL_INFO(ring->dev)->gen < 8) |
546 | flags |= (MI_SAVE_EXT_STATE_EN | MI_RESTORE_EXT_STATE_EN); |
537 | flags |= (MI_SAVE_EXT_STATE_EN | MI_RESTORE_EXT_STATE_EN); |
547 | 538 | ||
548 | 539 | ||
549 | len = 4; |
540 | len = 4; |
550 | if (INTEL_INFO(ring->dev)->gen >= 7) |
541 | if (INTEL_INFO(ring->dev)->gen >= 7) |
551 | len += 2 + (num_rings ? 4*num_rings + 2 : 0); |
542 | len += 2 + (num_rings ? 4*num_rings + 2 : 0); |
552 | 543 | ||
553 | ret = intel_ring_begin(req, len); |
544 | ret = intel_ring_begin(req, len); |
554 | if (ret) |
545 | if (ret) |
555 | return ret; |
546 | return ret; |
556 | 547 | ||
557 | /* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw,bdw,chv */ |
548 | /* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw,bdw,chv */ |
558 | if (INTEL_INFO(ring->dev)->gen >= 7) { |
549 | if (INTEL_INFO(ring->dev)->gen >= 7) { |
559 | intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_DISABLE); |
550 | intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_DISABLE); |
560 | if (num_rings) { |
551 | if (num_rings) { |
561 | struct intel_engine_cs *signaller; |
552 | struct intel_engine_cs *signaller; |
562 | 553 | ||
563 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(num_rings)); |
554 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(num_rings)); |
564 | for_each_ring(signaller, to_i915(ring->dev), i) { |
555 | for_each_ring(signaller, to_i915(ring->dev), i) { |
565 | if (signaller == ring) |
556 | if (signaller == ring) |
566 | continue; |
557 | continue; |
567 | 558 | ||
568 | intel_ring_emit_reg(ring, RING_PSMI_CTL(signaller->mmio_base)); |
559 | intel_ring_emit_reg(ring, RING_PSMI_CTL(signaller->mmio_base)); |
569 | intel_ring_emit(ring, _MASKED_BIT_ENABLE(GEN6_PSMI_SLEEP_MSG_DISABLE)); |
560 | intel_ring_emit(ring, _MASKED_BIT_ENABLE(GEN6_PSMI_SLEEP_MSG_DISABLE)); |
570 | } |
561 | } |
571 | } |
562 | } |
572 | } |
563 | } |
573 | 564 | ||
574 | intel_ring_emit(ring, MI_NOOP); |
565 | intel_ring_emit(ring, MI_NOOP); |
575 | intel_ring_emit(ring, MI_SET_CONTEXT); |
566 | intel_ring_emit(ring, MI_SET_CONTEXT); |
576 | intel_ring_emit(ring, i915_gem_obj_ggtt_offset(req->ctx->legacy_hw_ctx.rcs_state) | |
567 | intel_ring_emit(ring, i915_gem_obj_ggtt_offset(req->ctx->legacy_hw_ctx.rcs_state) | |
577 | flags); |
568 | flags); |
578 | /* |
569 | /* |
579 | * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP |
570 | * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP |
580 | * WaMiSetContext_Hang:snb,ivb,vlv |
571 | * WaMiSetContext_Hang:snb,ivb,vlv |
581 | */ |
572 | */ |
582 | intel_ring_emit(ring, MI_NOOP); |
573 | intel_ring_emit(ring, MI_NOOP); |
583 | 574 | ||
584 | if (INTEL_INFO(ring->dev)->gen >= 7) { |
575 | if (INTEL_INFO(ring->dev)->gen >= 7) { |
585 | if (num_rings) { |
576 | if (num_rings) { |
586 | struct intel_engine_cs *signaller; |
577 | struct intel_engine_cs *signaller; |
587 | 578 | ||
588 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(num_rings)); |
579 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(num_rings)); |
589 | for_each_ring(signaller, to_i915(ring->dev), i) { |
580 | for_each_ring(signaller, to_i915(ring->dev), i) { |
590 | if (signaller == ring) |
581 | if (signaller == ring) |
591 | continue; |
582 | continue; |
592 | 583 | ||
593 | intel_ring_emit_reg(ring, RING_PSMI_CTL(signaller->mmio_base)); |
584 | intel_ring_emit_reg(ring, RING_PSMI_CTL(signaller->mmio_base)); |
594 | intel_ring_emit(ring, _MASKED_BIT_DISABLE(GEN6_PSMI_SLEEP_MSG_DISABLE)); |
585 | intel_ring_emit(ring, _MASKED_BIT_DISABLE(GEN6_PSMI_SLEEP_MSG_DISABLE)); |
595 | } |
586 | } |
596 | } |
587 | } |
597 | intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_ENABLE); |
588 | intel_ring_emit(ring, MI_ARB_ON_OFF | MI_ARB_ENABLE); |
598 | } |
589 | } |
599 | 590 | ||
600 | intel_ring_advance(ring); |
591 | intel_ring_advance(ring); |
601 | 592 | ||
602 | return ret; |
593 | return ret; |
603 | } |
594 | } |
604 | 595 | ||
605 | static inline bool should_skip_switch(struct intel_engine_cs *ring, |
596 | static inline bool should_skip_switch(struct intel_engine_cs *ring, |
606 | struct intel_context *from, |
597 | struct intel_context *from, |
607 | struct intel_context *to) |
598 | struct intel_context *to) |
608 | { |
599 | { |
609 | if (to->remap_slice) |
600 | if (to->remap_slice) |
610 | return false; |
601 | return false; |
611 | 602 | ||
612 | if (to->ppgtt && from == to && |
603 | if (to->ppgtt && from == to && |
613 | !(intel_ring_flag(ring) & to->ppgtt->pd_dirty_rings)) |
604 | !(intel_ring_flag(ring) & to->ppgtt->pd_dirty_rings)) |
614 | return true; |
605 | return true; |
615 | 606 | ||
616 | return false; |
607 | return false; |
617 | } |
608 | } |
618 | 609 | ||
619 | static bool |
610 | static bool |
620 | needs_pd_load_pre(struct intel_engine_cs *ring, struct intel_context *to) |
611 | needs_pd_load_pre(struct intel_engine_cs *ring, struct intel_context *to) |
621 | { |
612 | { |
622 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
613 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
623 | 614 | ||
624 | if (!to->ppgtt) |
615 | if (!to->ppgtt) |
625 | return false; |
616 | return false; |
626 | 617 | ||
627 | if (INTEL_INFO(ring->dev)->gen < 8) |
618 | if (INTEL_INFO(ring->dev)->gen < 8) |
628 | return true; |
619 | return true; |
629 | 620 | ||
630 | if (ring != &dev_priv->ring[RCS]) |
621 | if (ring != &dev_priv->ring[RCS]) |
631 | return true; |
622 | return true; |
632 | 623 | ||
633 | return false; |
624 | return false; |
634 | } |
625 | } |
635 | 626 | ||
636 | static bool |
627 | static bool |
637 | needs_pd_load_post(struct intel_engine_cs *ring, struct intel_context *to, |
628 | needs_pd_load_post(struct intel_engine_cs *ring, struct intel_context *to, |
638 | u32 hw_flags) |
629 | u32 hw_flags) |
639 | { |
630 | { |
640 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
631 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
641 | 632 | ||
642 | if (!to->ppgtt) |
633 | if (!to->ppgtt) |
643 | return false; |
634 | return false; |
644 | 635 | ||
645 | if (!IS_GEN8(ring->dev)) |
636 | if (!IS_GEN8(ring->dev)) |
646 | return false; |
637 | return false; |
647 | 638 | ||
648 | if (ring != &dev_priv->ring[RCS]) |
639 | if (ring != &dev_priv->ring[RCS]) |
649 | return false; |
640 | return false; |
650 | 641 | ||
651 | if (hw_flags & MI_RESTORE_INHIBIT) |
642 | if (hw_flags & MI_RESTORE_INHIBIT) |
652 | return true; |
643 | return true; |
653 | 644 | ||
654 | return false; |
645 | return false; |
655 | } |
646 | } |
656 | 647 | ||
657 | static int do_switch(struct drm_i915_gem_request *req) |
648 | static int do_switch(struct drm_i915_gem_request *req) |
658 | { |
649 | { |
659 | struct intel_context *to = req->ctx; |
650 | struct intel_context *to = req->ctx; |
660 | struct intel_engine_cs *ring = req->ring; |
651 | struct intel_engine_cs *ring = req->ring; |
661 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
652 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
662 | struct intel_context *from = ring->last_context; |
653 | struct intel_context *from = ring->last_context; |
663 | u32 hw_flags = 0; |
654 | u32 hw_flags = 0; |
664 | bool uninitialized = false; |
655 | bool uninitialized = false; |
665 | int ret, i; |
656 | int ret, i; |
666 | 657 | ||
667 | if (from != NULL && ring == &dev_priv->ring[RCS]) { |
658 | if (from != NULL && ring == &dev_priv->ring[RCS]) { |
668 | BUG_ON(from->legacy_hw_ctx.rcs_state == NULL); |
659 | BUG_ON(from->legacy_hw_ctx.rcs_state == NULL); |
669 | BUG_ON(!i915_gem_obj_is_pinned(from->legacy_hw_ctx.rcs_state)); |
660 | BUG_ON(!i915_gem_obj_is_pinned(from->legacy_hw_ctx.rcs_state)); |
670 | } |
661 | } |
671 | 662 | ||
672 | if (should_skip_switch(ring, from, to)) |
663 | if (should_skip_switch(ring, from, to)) |
673 | return 0; |
664 | return 0; |
674 | 665 | ||
675 | /* Trying to pin first makes error handling easier. */ |
666 | /* Trying to pin first makes error handling easier. */ |
676 | if (ring == &dev_priv->ring[RCS]) { |
667 | if (ring == &dev_priv->ring[RCS]) { |
677 | ret = i915_gem_obj_ggtt_pin(to->legacy_hw_ctx.rcs_state, |
668 | ret = i915_gem_obj_ggtt_pin(to->legacy_hw_ctx.rcs_state, |
678 | get_context_alignment(ring->dev), 0); |
669 | get_context_alignment(ring->dev), 0); |
679 | if (ret) |
670 | if (ret) |
680 | return ret; |
671 | return ret; |
681 | } |
672 | } |
682 | 673 | ||
683 | /* |
674 | /* |
684 | * Pin can switch back to the default context if we end up calling into |
675 | * Pin can switch back to the default context if we end up calling into |
685 | * evict_everything - as a last ditch gtt defrag effort that also |
676 | * evict_everything - as a last ditch gtt defrag effort that also |
686 | * switches to the default context. Hence we need to reload from here. |
677 | * switches to the default context. Hence we need to reload from here. |
687 | */ |
678 | */ |
688 | from = ring->last_context; |
679 | from = ring->last_context; |
689 | 680 | ||
690 | if (needs_pd_load_pre(ring, to)) { |
681 | if (needs_pd_load_pre(ring, to)) { |
691 | /* Older GENs and non render rings still want the load first, |
682 | /* Older GENs and non render rings still want the load first, |
692 | * "PP_DCLV followed by PP_DIR_BASE register through Load |
683 | * "PP_DCLV followed by PP_DIR_BASE register through Load |
693 | * Register Immediate commands in Ring Buffer before submitting |
684 | * Register Immediate commands in Ring Buffer before submitting |
694 | * a context."*/ |
685 | * a context."*/ |
695 | trace_switch_mm(ring, to); |
686 | trace_switch_mm(ring, to); |
696 | ret = to->ppgtt->switch_mm(to->ppgtt, req); |
687 | ret = to->ppgtt->switch_mm(to->ppgtt, req); |
697 | if (ret) |
688 | if (ret) |
698 | goto unpin_out; |
689 | goto unpin_out; |
699 | 690 | ||
700 | /* Doing a PD load always reloads the page dirs */ |
691 | /* Doing a PD load always reloads the page dirs */ |
701 | to->ppgtt->pd_dirty_rings &= ~intel_ring_flag(ring); |
692 | to->ppgtt->pd_dirty_rings &= ~intel_ring_flag(ring); |
702 | } |
693 | } |
703 | 694 | ||
704 | if (ring != &dev_priv->ring[RCS]) { |
695 | if (ring != &dev_priv->ring[RCS]) { |
705 | if (from) |
696 | if (from) |
706 | i915_gem_context_unreference(from); |
697 | i915_gem_context_unreference(from); |
707 | goto done; |
698 | goto done; |
708 | } |
699 | } |
709 | 700 | ||
710 | /* |
701 | /* |
711 | * Clear this page out of any CPU caches for coherent swap-in/out. Note |
702 | * Clear this page out of any CPU caches for coherent swap-in/out. Note |
712 | * that thanks to write = false in this call and us not setting any gpu |
703 | * that thanks to write = false in this call and us not setting any gpu |
713 | * write domains when putting a context object onto the active list |
704 | * write domains when putting a context object onto the active list |
714 | * (when switching away from it), this won't block. |
705 | * (when switching away from it), this won't block. |
715 | * |
706 | * |
716 | * XXX: We need a real interface to do this instead of trickery. |
707 | * XXX: We need a real interface to do this instead of trickery. |
717 | */ |
708 | */ |
718 | ret = i915_gem_object_set_to_gtt_domain(to->legacy_hw_ctx.rcs_state, false); |
709 | ret = i915_gem_object_set_to_gtt_domain(to->legacy_hw_ctx.rcs_state, false); |
719 | if (ret) |
710 | if (ret) |
720 | goto unpin_out; |
711 | goto unpin_out; |
721 | 712 | ||
722 | if (!to->legacy_hw_ctx.initialized || i915_gem_context_is_default(to)) { |
713 | if (!to->legacy_hw_ctx.initialized || i915_gem_context_is_default(to)) { |
723 | hw_flags |= MI_RESTORE_INHIBIT; |
714 | hw_flags |= MI_RESTORE_INHIBIT; |
724 | /* NB: If we inhibit the restore, the context is not allowed to |
715 | /* NB: If we inhibit the restore, the context is not allowed to |
725 | * die because future work may end up depending on valid address |
716 | * die because future work may end up depending on valid address |
726 | * space. This means we must enforce that a page table load |
717 | * space. This means we must enforce that a page table load |
727 | * occur when this occurs. */ |
718 | * occur when this occurs. */ |
728 | } else if (to->ppgtt && |
719 | } else if (to->ppgtt && |
729 | (intel_ring_flag(ring) & to->ppgtt->pd_dirty_rings)) { |
720 | (intel_ring_flag(ring) & to->ppgtt->pd_dirty_rings)) { |
730 | hw_flags |= MI_FORCE_RESTORE; |
721 | hw_flags |= MI_FORCE_RESTORE; |
731 | to->ppgtt->pd_dirty_rings &= ~intel_ring_flag(ring); |
722 | to->ppgtt->pd_dirty_rings &= ~intel_ring_flag(ring); |
732 | } |
723 | } |
733 | 724 | ||
734 | /* We should never emit switch_mm more than once */ |
725 | /* We should never emit switch_mm more than once */ |
735 | WARN_ON(needs_pd_load_pre(ring, to) && |
726 | WARN_ON(needs_pd_load_pre(ring, to) && |
736 | needs_pd_load_post(ring, to, hw_flags)); |
727 | needs_pd_load_post(ring, to, hw_flags)); |
737 | 728 | ||
738 | ret = mi_set_context(req, hw_flags); |
729 | ret = mi_set_context(req, hw_flags); |
739 | if (ret) |
730 | if (ret) |
740 | goto unpin_out; |
731 | goto unpin_out; |
741 | 732 | ||
742 | /* GEN8 does *not* require an explicit reload if the PDPs have been |
733 | /* GEN8 does *not* require an explicit reload if the PDPs have been |
743 | * setup, and we do not wish to move them. |
734 | * setup, and we do not wish to move them. |
744 | */ |
735 | */ |
745 | if (needs_pd_load_post(ring, to, hw_flags)) { |
736 | if (needs_pd_load_post(ring, to, hw_flags)) { |
746 | trace_switch_mm(ring, to); |
737 | trace_switch_mm(ring, to); |
747 | ret = to->ppgtt->switch_mm(to->ppgtt, req); |
738 | ret = to->ppgtt->switch_mm(to->ppgtt, req); |
748 | /* The hardware context switch is emitted, but we haven't |
739 | /* The hardware context switch is emitted, but we haven't |
749 | * actually changed the state - so it's probably safe to bail |
740 | * actually changed the state - so it's probably safe to bail |
750 | * here. Still, let the user know something dangerous has |
741 | * here. Still, let the user know something dangerous has |
751 | * happened. |
742 | * happened. |
752 | */ |
743 | */ |
753 | if (ret) { |
744 | if (ret) { |
754 | DRM_ERROR("Failed to change address space on context switch\n"); |
745 | DRM_ERROR("Failed to change address space on context switch\n"); |
755 | goto unpin_out; |
746 | goto unpin_out; |
756 | } |
747 | } |
757 | } |
748 | } |
758 | 749 | ||
759 | for (i = 0; i < MAX_L3_SLICES; i++) { |
750 | for (i = 0; i < MAX_L3_SLICES; i++) { |
760 | if (!(to->remap_slice & (1< |
751 | if (!(to->remap_slice & (1< |
761 | continue; |
752 | continue; |
762 | 753 | ||
763 | ret = i915_gem_l3_remap(req, i); |
754 | ret = i915_gem_l3_remap(req, i); |
764 | /* If it failed, try again next round */ |
755 | /* If it failed, try again next round */ |
765 | if (ret) |
756 | if (ret) |
766 | DRM_DEBUG_DRIVER("L3 remapping failed\n"); |
757 | DRM_DEBUG_DRIVER("L3 remapping failed\n"); |
767 | else |
758 | else |
768 | to->remap_slice &= ~(1< |
759 | to->remap_slice &= ~(1< |
769 | } |
760 | } |
770 | 761 | ||
771 | /* The backing object for the context is done after switching to the |
762 | /* The backing object for the context is done after switching to the |
772 | * *next* context. Therefore we cannot retire the previous context until |
763 | * *next* context. Therefore we cannot retire the previous context until |
773 | * the next context has already started running. In fact, the below code |
764 | * the next context has already started running. In fact, the below code |
774 | * is a bit suboptimal because the retiring can occur simply after the |
765 | * is a bit suboptimal because the retiring can occur simply after the |
775 | * MI_SET_CONTEXT instead of when the next seqno has completed. |
766 | * MI_SET_CONTEXT instead of when the next seqno has completed. |
776 | */ |
767 | */ |
777 | if (from != NULL) { |
768 | if (from != NULL) { |
778 | from->legacy_hw_ctx.rcs_state->base.read_domains = I915_GEM_DOMAIN_INSTRUCTION; |
769 | from->legacy_hw_ctx.rcs_state->base.read_domains = I915_GEM_DOMAIN_INSTRUCTION; |
779 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(from->legacy_hw_ctx.rcs_state), req); |
770 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(from->legacy_hw_ctx.rcs_state), req); |
780 | /* As long as MI_SET_CONTEXT is serializing, ie. it flushes the |
771 | /* As long as MI_SET_CONTEXT is serializing, ie. it flushes the |
781 | * whole damn pipeline, we don't need to explicitly mark the |
772 | * whole damn pipeline, we don't need to explicitly mark the |
782 | * object dirty. The only exception is that the context must be |
773 | * object dirty. The only exception is that the context must be |
783 | * correct in case the object gets swapped out. Ideally we'd be |
774 | * correct in case the object gets swapped out. Ideally we'd be |
784 | * able to defer doing this until we know the object would be |
775 | * able to defer doing this until we know the object would be |
785 | * swapped, but there is no way to do that yet. |
776 | * swapped, but there is no way to do that yet. |
786 | */ |
777 | */ |
787 | from->legacy_hw_ctx.rcs_state->dirty = 1; |
778 | from->legacy_hw_ctx.rcs_state->dirty = 1; |
788 | 779 | ||
789 | /* obj is kept alive until the next request by its active ref */ |
780 | /* obj is kept alive until the next request by its active ref */ |
790 | i915_gem_object_ggtt_unpin(from->legacy_hw_ctx.rcs_state); |
781 | i915_gem_object_ggtt_unpin(from->legacy_hw_ctx.rcs_state); |
791 | i915_gem_context_unreference(from); |
782 | i915_gem_context_unreference(from); |
792 | } |
783 | } |
793 | 784 | ||
794 | uninitialized = !to->legacy_hw_ctx.initialized; |
785 | uninitialized = !to->legacy_hw_ctx.initialized; |
795 | to->legacy_hw_ctx.initialized = true; |
786 | to->legacy_hw_ctx.initialized = true; |
796 | 787 | ||
797 | done: |
788 | done: |
798 | i915_gem_context_reference(to); |
789 | i915_gem_context_reference(to); |
799 | ring->last_context = to; |
790 | ring->last_context = to; |
800 | 791 | ||
801 | if (uninitialized) { |
792 | if (uninitialized) { |
802 | if (ring->init_context) { |
793 | if (ring->init_context) { |
803 | ret = ring->init_context(req); |
794 | ret = ring->init_context(req); |
804 | if (ret) |
795 | if (ret) |
805 | DRM_ERROR("ring init context: %d\n", ret); |
796 | DRM_ERROR("ring init context: %d\n", ret); |
806 | } |
797 | } |
807 | } |
798 | } |
808 | 799 | ||
809 | return 0; |
800 | return 0; |
810 | 801 | ||
811 | unpin_out: |
802 | unpin_out: |
812 | if (ring->id == RCS) |
803 | if (ring->id == RCS) |
813 | i915_gem_object_ggtt_unpin(to->legacy_hw_ctx.rcs_state); |
804 | i915_gem_object_ggtt_unpin(to->legacy_hw_ctx.rcs_state); |
814 | return ret; |
805 | return ret; |
815 | } |
806 | } |
816 | 807 | ||
817 | /** |
808 | /** |
818 | * i915_switch_context() - perform a GPU context switch. |
809 | * i915_switch_context() - perform a GPU context switch. |
819 | * @req: request for which we'll execute the context switch |
810 | * @req: request for which we'll execute the context switch |
820 | * |
811 | * |
821 | * The context life cycle is simple. The context refcount is incremented and |
812 | * The context life cycle is simple. The context refcount is incremented and |
822 | * decremented by 1 and create and destroy. If the context is in use by the GPU, |
813 | * decremented by 1 and create and destroy. If the context is in use by the GPU, |
823 | * it will have a refcount > 1. This allows us to destroy the context abstract |
814 | * it will have a refcount > 1. This allows us to destroy the context abstract |
824 | * object while letting the normal object tracking destroy the backing BO. |
815 | * object while letting the normal object tracking destroy the backing BO. |
825 | * |
816 | * |
826 | * This function should not be used in execlists mode. Instead the context is |
817 | * This function should not be used in execlists mode. Instead the context is |
827 | * switched by writing to the ELSP and requests keep a reference to their |
818 | * switched by writing to the ELSP and requests keep a reference to their |
828 | * context. |
819 | * context. |
829 | */ |
820 | */ |
830 | int i915_switch_context(struct drm_i915_gem_request *req) |
821 | int i915_switch_context(struct drm_i915_gem_request *req) |
831 | { |
822 | { |
832 | struct intel_engine_cs *ring = req->ring; |
823 | struct intel_engine_cs *ring = req->ring; |
833 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
824 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
834 | 825 | ||
835 | WARN_ON(i915.enable_execlists); |
826 | WARN_ON(i915.enable_execlists); |
836 | WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex)); |
827 | WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex)); |
837 | 828 | ||
838 | if (req->ctx->legacy_hw_ctx.rcs_state == NULL) { /* We have the fake context */ |
829 | if (req->ctx->legacy_hw_ctx.rcs_state == NULL) { /* We have the fake context */ |
839 | if (req->ctx != ring->last_context) { |
830 | if (req->ctx != ring->last_context) { |
840 | i915_gem_context_reference(req->ctx); |
831 | i915_gem_context_reference(req->ctx); |
841 | if (ring->last_context) |
832 | if (ring->last_context) |
842 | i915_gem_context_unreference(ring->last_context); |
833 | i915_gem_context_unreference(ring->last_context); |
843 | ring->last_context = req->ctx; |
834 | ring->last_context = req->ctx; |
844 | } |
835 | } |
845 | return 0; |
836 | return 0; |
846 | } |
837 | } |
847 | 838 | ||
848 | return do_switch(req); |
839 | return do_switch(req); |
849 | } |
840 | } |
850 | 841 | ||
851 | static bool contexts_enabled(struct drm_device *dev) |
842 | static bool contexts_enabled(struct drm_device *dev) |
852 | { |
843 | { |
853 | return i915.enable_execlists || to_i915(dev)->hw_context_size; |
844 | return i915.enable_execlists || to_i915(dev)->hw_context_size; |
854 | } |
845 | } |
855 | 846 | ||
856 | int i915_gem_context_create_ioctl(struct drm_device *dev, void *data, |
847 | int i915_gem_context_create_ioctl(struct drm_device *dev, void *data, |
857 | struct drm_file *file) |
848 | struct drm_file *file) |
858 | { |
849 | { |
859 | struct drm_i915_gem_context_create *args = data; |
850 | struct drm_i915_gem_context_create *args = data; |
860 | struct drm_i915_file_private *file_priv = file->driver_priv; |
851 | struct drm_i915_file_private *file_priv = file->driver_priv; |
861 | struct intel_context *ctx; |
852 | struct intel_context *ctx; |
862 | int ret; |
853 | int ret; |
863 | 854 | ||
864 | if (!contexts_enabled(dev)) |
855 | if (!contexts_enabled(dev)) |
865 | return -ENODEV; |
856 | return -ENODEV; |
- | 857 | ||
- | 858 | if (args->pad != 0) |
|
- | 859 | return -EINVAL; |
|
866 | 860 | ||
867 | ret = i915_mutex_lock_interruptible(dev); |
861 | ret = i915_mutex_lock_interruptible(dev); |
868 | if (ret) |
862 | if (ret) |
869 | return ret; |
863 | return ret; |
870 | 864 | ||
871 | ctx = i915_gem_create_context(dev, file_priv); |
865 | ctx = i915_gem_create_context(dev, file_priv); |
872 | mutex_unlock(&dev->struct_mutex); |
866 | mutex_unlock(&dev->struct_mutex); |
873 | if (IS_ERR(ctx)) |
867 | if (IS_ERR(ctx)) |
874 | return PTR_ERR(ctx); |
868 | return PTR_ERR(ctx); |
875 | 869 | ||
876 | args->ctx_id = ctx->user_handle; |
870 | args->ctx_id = ctx->user_handle; |
877 | DRM_DEBUG_DRIVER("HW context %d created\n", args->ctx_id); |
871 | DRM_DEBUG_DRIVER("HW context %d created\n", args->ctx_id); |
878 | 872 | ||
879 | return 0; |
873 | return 0; |
880 | } |
874 | } |
881 | 875 | ||
882 | int i915_gem_context_destroy_ioctl(struct drm_device *dev, void *data, |
876 | int i915_gem_context_destroy_ioctl(struct drm_device *dev, void *data, |
883 | struct drm_file *file) |
877 | struct drm_file *file) |
884 | { |
878 | { |
885 | struct drm_i915_gem_context_destroy *args = data; |
879 | struct drm_i915_gem_context_destroy *args = data; |
886 | struct drm_i915_file_private *file_priv = file->driver_priv; |
880 | struct drm_i915_file_private *file_priv = file->driver_priv; |
887 | struct intel_context *ctx; |
881 | struct intel_context *ctx; |
888 | int ret; |
882 | int ret; |
- | 883 | ||
- | 884 | if (args->pad != 0) |
|
- | 885 | return -EINVAL; |
|
889 | 886 | ||
890 | if (args->ctx_id == DEFAULT_CONTEXT_HANDLE) |
887 | if (args->ctx_id == DEFAULT_CONTEXT_HANDLE) |
891 | return -ENOENT; |
888 | return -ENOENT; |
892 | 889 | ||
893 | ret = i915_mutex_lock_interruptible(dev); |
890 | ret = i915_mutex_lock_interruptible(dev); |
894 | if (ret) |
891 | if (ret) |
895 | return ret; |
892 | return ret; |
896 | 893 | ||
897 | ctx = i915_gem_context_get(file_priv, args->ctx_id); |
894 | ctx = i915_gem_context_get(file_priv, args->ctx_id); |
898 | if (IS_ERR(ctx)) { |
895 | if (IS_ERR(ctx)) { |
899 | mutex_unlock(&dev->struct_mutex); |
896 | mutex_unlock(&dev->struct_mutex); |
900 | return PTR_ERR(ctx); |
897 | return PTR_ERR(ctx); |
901 | } |
898 | } |
902 | 899 | ||
903 | idr_remove(&ctx->file_priv->context_idr, ctx->user_handle); |
900 | idr_remove(&ctx->file_priv->context_idr, ctx->user_handle); |
904 | i915_gem_context_unreference(ctx); |
901 | i915_gem_context_unreference(ctx); |
905 | mutex_unlock(&dev->struct_mutex); |
902 | mutex_unlock(&dev->struct_mutex); |
906 | 903 | ||
907 | DRM_DEBUG_DRIVER("HW context %d destroyed\n", args->ctx_id); |
904 | DRM_DEBUG_DRIVER("HW context %d destroyed\n", args->ctx_id); |
908 | return 0; |
905 | return 0; |
909 | } |
906 | } |
910 | 907 | ||
911 | int i915_gem_context_getparam_ioctl(struct drm_device *dev, void *data, |
908 | int i915_gem_context_getparam_ioctl(struct drm_device *dev, void *data, |
912 | struct drm_file *file) |
909 | struct drm_file *file) |
913 | { |
910 | { |
914 | struct drm_i915_file_private *file_priv = file->driver_priv; |
911 | struct drm_i915_file_private *file_priv = file->driver_priv; |
915 | struct drm_i915_gem_context_param *args = data; |
912 | struct drm_i915_gem_context_param *args = data; |
916 | struct intel_context *ctx; |
913 | struct intel_context *ctx; |
917 | int ret; |
914 | int ret; |
918 | 915 | ||
919 | ret = i915_mutex_lock_interruptible(dev); |
916 | ret = i915_mutex_lock_interruptible(dev); |
920 | if (ret) |
917 | if (ret) |
921 | return ret; |
918 | return ret; |
922 | 919 | ||
923 | ctx = i915_gem_context_get(file_priv, args->ctx_id); |
920 | ctx = i915_gem_context_get(file_priv, args->ctx_id); |
924 | if (IS_ERR(ctx)) { |
921 | if (IS_ERR(ctx)) { |
925 | mutex_unlock(&dev->struct_mutex); |
922 | mutex_unlock(&dev->struct_mutex); |
926 | return PTR_ERR(ctx); |
923 | return PTR_ERR(ctx); |
927 | } |
924 | } |
928 | 925 | ||
929 | args->size = 0; |
926 | args->size = 0; |
930 | switch (args->param) { |
927 | switch (args->param) { |
931 | case I915_CONTEXT_PARAM_BAN_PERIOD: |
928 | case I915_CONTEXT_PARAM_BAN_PERIOD: |
932 | args->value = ctx->hang_stats.ban_period_seconds; |
929 | args->value = ctx->hang_stats.ban_period_seconds; |
933 | break; |
930 | break; |
934 | case I915_CONTEXT_PARAM_NO_ZEROMAP: |
931 | case I915_CONTEXT_PARAM_NO_ZEROMAP: |
935 | args->value = ctx->flags & CONTEXT_NO_ZEROMAP; |
932 | args->value = ctx->flags & CONTEXT_NO_ZEROMAP; |
936 | break; |
933 | break; |
937 | case I915_CONTEXT_PARAM_GTT_SIZE: |
934 | case I915_CONTEXT_PARAM_GTT_SIZE: |
938 | if (ctx->ppgtt) |
935 | if (ctx->ppgtt) |
939 | args->value = ctx->ppgtt->base.total; |
936 | args->value = ctx->ppgtt->base.total; |
940 | else if (to_i915(dev)->mm.aliasing_ppgtt) |
937 | else if (to_i915(dev)->mm.aliasing_ppgtt) |
941 | args->value = to_i915(dev)->mm.aliasing_ppgtt->base.total; |
938 | args->value = to_i915(dev)->mm.aliasing_ppgtt->base.total; |
942 | else |
939 | else |
943 | args->value = to_i915(dev)->gtt.base.total; |
940 | args->value = to_i915(dev)->gtt.base.total; |
944 | break; |
941 | break; |
945 | default: |
942 | default: |
946 | ret = -EINVAL; |
943 | ret = -EINVAL; |
947 | break; |
944 | break; |
948 | } |
945 | } |
949 | mutex_unlock(&dev->struct_mutex); |
946 | mutex_unlock(&dev->struct_mutex); |
950 | 947 | ||
951 | return ret; |
948 | return ret; |
952 | } |
949 | } |
953 | 950 | ||
954 | int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data, |
951 | int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data, |
955 | struct drm_file *file) |
952 | struct drm_file *file) |
956 | { |
953 | { |
957 | struct drm_i915_file_private *file_priv = file->driver_priv; |
954 | struct drm_i915_file_private *file_priv = file->driver_priv; |
958 | struct drm_i915_gem_context_param *args = data; |
955 | struct drm_i915_gem_context_param *args = data; |
959 | struct intel_context *ctx; |
956 | struct intel_context *ctx; |
960 | int ret; |
957 | int ret; |
961 | 958 | ||
962 | ret = i915_mutex_lock_interruptible(dev); |
959 | ret = i915_mutex_lock_interruptible(dev); |
963 | if (ret) |
960 | if (ret) |
964 | return ret; |
961 | return ret; |
965 | 962 | ||
966 | ctx = i915_gem_context_get(file_priv, args->ctx_id); |
963 | ctx = i915_gem_context_get(file_priv, args->ctx_id); |
967 | if (IS_ERR(ctx)) { |
964 | if (IS_ERR(ctx)) { |
968 | mutex_unlock(&dev->struct_mutex); |
965 | mutex_unlock(&dev->struct_mutex); |
969 | return PTR_ERR(ctx); |
966 | return PTR_ERR(ctx); |
970 | } |
967 | } |
971 | 968 | ||
972 | switch (args->param) { |
969 | switch (args->param) { |
973 | case I915_CONTEXT_PARAM_BAN_PERIOD: |
970 | case I915_CONTEXT_PARAM_BAN_PERIOD: |
974 | if (args->size) |
971 | if (args->size) |
975 | ret = -EINVAL; |
972 | ret = -EINVAL; |
976 | else if (args->value < ctx->hang_stats.ban_period_seconds && |
973 | else if (args->value < ctx->hang_stats.ban_period_seconds && |
977 | !capable(CAP_SYS_ADMIN)) |
974 | !capable(CAP_SYS_ADMIN)) |
978 | ret = -EPERM; |
975 | ret = -EPERM; |
979 | else |
976 | else |
980 | ctx->hang_stats.ban_period_seconds = args->value; |
977 | ctx->hang_stats.ban_period_seconds = args->value; |
981 | break; |
978 | break; |
982 | case I915_CONTEXT_PARAM_NO_ZEROMAP: |
979 | case I915_CONTEXT_PARAM_NO_ZEROMAP: |
983 | if (args->size) { |
980 | if (args->size) { |
984 | ret = -EINVAL; |
981 | ret = -EINVAL; |
985 | } else { |
982 | } else { |
986 | ctx->flags &= ~CONTEXT_NO_ZEROMAP; |
983 | ctx->flags &= ~CONTEXT_NO_ZEROMAP; |
987 | ctx->flags |= args->value ? CONTEXT_NO_ZEROMAP : 0; |
984 | ctx->flags |= args->value ? CONTEXT_NO_ZEROMAP : 0; |
988 | } |
985 | } |
989 | break; |
986 | break; |
990 | default: |
987 | default: |
991 | ret = -EINVAL; |
988 | ret = -EINVAL; |
992 | break; |
989 | break; |
993 | } |
990 | } |
994 | mutex_unlock(&dev->struct_mutex); |
991 | mutex_unlock(&dev->struct_mutex); |
995 | 992 | ||
996 | return ret; |
993 | return ret; |
997 | }> |
994 | }> |
998 | > |
995 | > |
999 | > |
996 | > |
1000 | > |
997 | > |
1001 | >>>>>>20))><20))>>><>>10) |
998 | >>>>20))><20))>>><>>10) |
1002 | #define><10) |
999 | #define><10) |
1003 | #define> |
1000 | #define> |