Rev 1404 | Rev 1430 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 1404 | Rev 1428 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright 2008 Advanced Micro Devices, Inc. |
2 | * Copyright 2008 Advanced Micro Devices, Inc. |
3 | * Copyright 2008 Red Hat Inc. |
3 | * Copyright 2008 Red Hat Inc. |
4 | * Copyright 2009 Jerome Glisse. |
4 | * Copyright 2009 Jerome Glisse. |
5 | * |
5 | * |
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
7 | * copy of this software and associated documentation files (the "Software"), |
7 | * copy of this software and associated documentation files (the "Software"), |
8 | * to deal in the Software without restriction, including without limitation |
8 | * to deal in the Software without restriction, including without limitation |
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
10 | * and/or sell copies of the Software, and to permit persons to whom the |
10 | * and/or sell copies of the Software, and to permit persons to whom the |
11 | * Software is furnished to do so, subject to the following conditions: |
11 | * Software is furnished to do so, subject to the following conditions: |
12 | * |
12 | * |
13 | * The above copyright notice and this permission notice shall be included in |
13 | * The above copyright notice and this permission notice shall be included in |
14 | * all copies or substantial portions of the Software. |
14 | * all copies or substantial portions of the Software. |
15 | * |
15 | * |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
22 | * OTHER DEALINGS IN THE SOFTWARE. |
22 | * OTHER DEALINGS IN THE SOFTWARE. |
23 | * |
23 | * |
24 | * Authors: Dave Airlie |
24 | * Authors: Dave Airlie |
25 | * Alex Deucher |
25 | * Alex Deucher |
26 | * Jerome Glisse |
26 | * Jerome Glisse |
27 | */ |
27 | */ |
28 | #include |
28 | #include |
29 | #include "drmP.h" |
29 | #include "drmP.h" |
30 | #include "radeon_drm.h" |
30 | #include "radeon_drm.h" |
31 | #include "radeon_reg.h" |
31 | #include "radeon_reg.h" |
32 | #include "radeon.h" |
32 | #include "radeon.h" |
33 | #include "atom.h" |
33 | #include "atom.h" |
34 | 34 | ||
35 | int radeon_debugfs_ib_init(struct radeon_device *rdev); |
35 | int radeon_debugfs_ib_init(struct radeon_device *rdev); |
36 | 36 | ||
37 | /* |
37 | /* |
38 | * IB. |
38 | * IB. |
39 | */ |
39 | */ |
40 | 40 | ||
41 | #if 0 |
41 | #if 0 |
42 | 42 | ||
43 | int radeon_ib_get(struct radeon_device *rdev, struct radeon_ib **ib) |
43 | int radeon_ib_get(struct radeon_device *rdev, struct radeon_ib **ib) |
44 | { |
44 | { |
45 | struct radeon_fence *fence; |
45 | struct radeon_fence *fence; |
46 | struct radeon_ib *nib; |
46 | struct radeon_ib *nib; |
47 | unsigned long i; |
- | |
48 | int r = 0; |
47 | int r = 0, i, c; |
49 | 48 | ||
50 | *ib = NULL; |
49 | *ib = NULL; |
51 | r = radeon_fence_create(rdev, &fence); |
50 | r = radeon_fence_create(rdev, &fence); |
52 | if (r) { |
51 | if (r) { |
53 | DRM_ERROR("failed to create fence for new IB\n"); |
52 | dev_err(rdev->dev, "failed to create fence for new IB\n"); |
54 | return r; |
53 | return r; |
55 | } |
54 | } |
56 | mutex_lock(&rdev->ib_pool.mutex); |
55 | mutex_lock(&rdev->ib_pool.mutex); |
57 | i = find_first_zero_bit(rdev->ib_pool.alloc_bm, RADEON_IB_POOL_SIZE); |
56 | for (i = rdev->ib_pool.head_id, c = 0, nib = NULL; c < RADEON_IB_POOL_SIZE; c++, i++) { |
58 | if (i < RADEON_IB_POOL_SIZE) { |
57 | i &= (RADEON_IB_POOL_SIZE - 1); |
59 | set_bit(i, rdev->ib_pool.alloc_bm); |
- | |
60 | rdev->ib_pool.ibs[i].length_dw = 0; |
58 | if (rdev->ib_pool.ibs[i].free) { |
61 | *ib = &rdev->ib_pool.ibs[i]; |
59 | nib = &rdev->ib_pool.ibs[i]; |
62 | mutex_unlock(&rdev->ib_pool.mutex); |
- | |
63 | goto out; |
60 | break; |
64 | } |
61 | } |
- | 62 | } |
|
65 | if (list_empty(&rdev->ib_pool.scheduled_ibs)) { |
63 | if (nib == NULL) { |
66 | /* we go do nothings here */ |
64 | /* This should never happen, it means we allocated all |
67 | mutex_unlock(&rdev->ib_pool.mutex); |
65 | * IB and haven't scheduled one yet, return EBUSY to |
68 | DRM_ERROR("all IB allocated none scheduled.\n"); |
66 | * userspace hoping that on ioctl recall we get better |
69 | r = -EINVAL; |
- | |
70 | goto out; |
67 | * luck |
71 | } |
68 | */ |
72 | /* get the first ib on the scheduled list */ |
- | |
73 | nib = list_entry(rdev->ib_pool.scheduled_ibs.next, |
69 | dev_err(rdev->dev, "no free indirect buffer !\n"); |
74 | struct radeon_ib, list); |
- | |
75 | if (nib->fence == NULL) { |
- | |
76 | /* we go do nothings here */ |
- | |
77 | mutex_unlock(&rdev->ib_pool.mutex); |
70 | mutex_unlock(&rdev->ib_pool.mutex); |
78 | DRM_ERROR("IB %lu scheduled without a fence.\n", nib->idx); |
71 | radeon_fence_unref(&fence); |
79 | r = -EINVAL; |
72 | return -EBUSY; |
80 | goto out; |
- | |
81 | } |
73 | } |
- | 74 | rdev->ib_pool.head_id = (nib->idx + 1) & (RADEON_IB_POOL_SIZE - 1); |
|
- | 75 | nib->free = false; |
|
- | 76 | if (nib->fence) { |
|
82 | mutex_unlock(&rdev->ib_pool.mutex); |
77 | mutex_unlock(&rdev->ib_pool.mutex); |
83 | - | ||
84 | r = radeon_fence_wait(nib->fence, false); |
78 | r = radeon_fence_wait(nib->fence, false); |
85 | if (r) { |
79 | if (r) { |
86 | DRM_ERROR("radeon: IB(%lu:0x%016lX:%u)\n", nib->idx, |
80 | dev_err(rdev->dev, "error waiting fence of IB(%u:0x%016lX:%u)\n", |
87 | (unsigned long)nib->gpu_addr, nib->length_dw); |
81 | nib->idx, (unsigned long)nib->gpu_addr, nib->length_dw); |
- | 82 | mutex_lock(&rdev->ib_pool.mutex); |
|
- | 83 | nib->free = true; |
|
88 | DRM_ERROR("radeon: GPU lockup detected, fail to get a IB\n"); |
84 | mutex_unlock(&rdev->ib_pool.mutex); |
- | 85 | radeon_fence_unref(&fence); |
|
89 | goto out; |
86 | return r; |
- | 87 | } |
|
- | 88 | mutex_lock(&rdev->ib_pool.mutex); |
|
90 | } |
89 | } |
91 | radeon_fence_unref(&nib->fence); |
90 | radeon_fence_unref(&nib->fence); |
92 | - | ||
- | 91 | nib->fence = fence; |
|
93 | nib->length_dw = 0; |
92 | nib->length_dw = 0; |
94 | - | ||
95 | /* scheduled list is accessed here */ |
- | |
96 | mutex_lock(&rdev->ib_pool.mutex); |
- | |
97 | list_del(&nib->list); |
- | |
98 | INIT_LIST_HEAD(&nib->list); |
- | |
99 | mutex_unlock(&rdev->ib_pool.mutex); |
93 | mutex_unlock(&rdev->ib_pool.mutex); |
100 | - | ||
101 | *ib = nib; |
94 | *ib = nib; |
102 | out: |
- | |
103 | if (r) { |
- | |
104 | radeon_fence_unref(&fence); |
- | |
105 | } else { |
- | |
106 | (*ib)->fence = fence; |
- | |
107 | } |
- | |
108 | return r; |
95 | return 0; |
109 | } |
96 | } |
110 | 97 | ||
111 | void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib) |
98 | void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib) |
112 | { |
99 | { |
113 | struct radeon_ib *tmp = *ib; |
100 | struct radeon_ib *tmp = *ib; |
114 | 101 | ||
115 | *ib = NULL; |
102 | *ib = NULL; |
116 | if (tmp == NULL) { |
103 | if (tmp == NULL) { |
117 | return; |
104 | return; |
118 | } |
105 | } |
119 | mutex_lock(&rdev->ib_pool.mutex); |
- | |
120 | if (!list_empty(&tmp->list) && !radeon_fence_signaled(tmp->fence)) { |
- | |
121 | /* IB is scheduled & not signaled don't do anythings */ |
- | |
122 | mutex_unlock(&rdev->ib_pool.mutex); |
- | |
123 | return; |
- | |
124 | } |
- | |
125 | list_del(&tmp->list); |
- | |
126 | INIT_LIST_HEAD(&tmp->list); |
- | |
127 | if (tmp->fence) |
106 | if (!tmp->fence->emited) |
128 | radeon_fence_unref(&tmp->fence); |
107 | radeon_fence_unref(&tmp->fence); |
129 | - | ||
130 | tmp->length_dw = 0; |
108 | mutex_lock(&rdev->ib_pool.mutex); |
131 | clear_bit(tmp->idx, rdev->ib_pool.alloc_bm); |
109 | tmp->free = true; |
132 | mutex_unlock(&rdev->ib_pool.mutex); |
110 | mutex_unlock(&rdev->ib_pool.mutex); |
133 | } |
111 | } |
134 | 112 | ||
135 | int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib) |
113 | int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib) |
136 | { |
114 | { |
137 | int r = 0; |
115 | int r = 0; |
138 | 116 | ||
139 | if (!ib->length_dw || !rdev->cp.ready) { |
117 | if (!ib->length_dw || !rdev->cp.ready) { |
140 | /* TODO: Nothings in the ib we should report. */ |
118 | /* TODO: Nothings in the ib we should report. */ |
141 | DRM_ERROR("radeon: couldn't schedule IB(%lu).\n", ib->idx); |
119 | DRM_ERROR("radeon: couldn't schedule IB(%u).\n", ib->idx); |
142 | return -EINVAL; |
120 | return -EINVAL; |
143 | } |
121 | } |
144 | 122 | ||
145 | /* 64 dwords should be enough for fence too */ |
123 | /* 64 dwords should be enough for fence too */ |
146 | r = radeon_ring_lock(rdev, 64); |
124 | r = radeon_ring_lock(rdev, 64); |
147 | if (r) { |
125 | if (r) { |
148 | DRM_ERROR("radeon: scheduling IB failled (%d).\n", r); |
126 | DRM_ERROR("radeon: scheduling IB failled (%d).\n", r); |
149 | return r; |
127 | return r; |
150 | } |
128 | } |
151 | radeon_ring_ib_execute(rdev, ib); |
129 | radeon_ring_ib_execute(rdev, ib); |
152 | radeon_fence_emit(rdev, ib->fence); |
130 | radeon_fence_emit(rdev, ib->fence); |
153 | mutex_lock(&rdev->ib_pool.mutex); |
131 | mutex_lock(&rdev->ib_pool.mutex); |
154 | list_add_tail(&ib->list, &rdev->ib_pool.scheduled_ibs); |
132 | /* once scheduled IB is considered free and protected by the fence */ |
- | 133 | ib->free = true; |
|
155 | mutex_unlock(&rdev->ib_pool.mutex); |
134 | mutex_unlock(&rdev->ib_pool.mutex); |
156 | radeon_ring_unlock_commit(rdev); |
135 | radeon_ring_unlock_commit(rdev); |
157 | return 0; |
136 | return 0; |
158 | } |
137 | } |
159 | #endif |
138 | #endif |
160 | 139 | ||
161 | int radeon_ib_pool_init(struct radeon_device *rdev) |
140 | int radeon_ib_pool_init(struct radeon_device *rdev) |
162 | { |
141 | { |
163 | void *ptr; |
142 | void *ptr; |
164 | uint64_t gpu_addr; |
143 | uint64_t gpu_addr; |
165 | int i; |
144 | int i; |
166 | int r = 0; |
145 | int r = 0; |
167 | 146 | ||
168 | if (rdev->ib_pool.robj) |
147 | if (rdev->ib_pool.robj) |
169 | return 0; |
148 | return 0; |
170 | /* Allocate 1M object buffer */ |
149 | /* Allocate 1M object buffer */ |
171 | INIT_LIST_HEAD(&rdev->ib_pool.scheduled_ibs); |
- | |
172 | r = radeon_bo_create(rdev, NULL, RADEON_IB_POOL_SIZE*64*1024, |
150 | r = radeon_bo_create(rdev, NULL, RADEON_IB_POOL_SIZE*64*1024, |
173 | true, RADEON_GEM_DOMAIN_GTT, |
151 | true, RADEON_GEM_DOMAIN_GTT, |
174 | &rdev->ib_pool.robj); |
152 | &rdev->ib_pool.robj); |
175 | if (r) { |
153 | if (r) { |
176 | DRM_ERROR("radeon: failed to ib pool (%d).\n", r); |
154 | DRM_ERROR("radeon: failed to ib pool (%d).\n", r); |
177 | return r; |
155 | return r; |
178 | } |
156 | } |
179 | r = radeon_bo_reserve(rdev->ib_pool.robj, false); |
157 | r = radeon_bo_reserve(rdev->ib_pool.robj, false); |
180 | if (unlikely(r != 0)) |
158 | if (unlikely(r != 0)) |
181 | return r; |
159 | return r; |
182 | r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr); |
160 | r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr); |
183 | if (r) { |
161 | if (r) { |
184 | radeon_bo_unreserve(rdev->ib_pool.robj); |
162 | radeon_bo_unreserve(rdev->ib_pool.robj); |
185 | DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r); |
163 | DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r); |
186 | return r; |
164 | return r; |
187 | } |
165 | } |
188 | r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr); |
166 | r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr); |
189 | radeon_bo_unreserve(rdev->ib_pool.robj); |
167 | radeon_bo_unreserve(rdev->ib_pool.robj); |
190 | if (r) { |
168 | if (r) { |
191 | DRM_ERROR("radeon: failed to map ib poll (%d).\n", r); |
169 | DRM_ERROR("radeon: failed to map ib poll (%d).\n", r); |
192 | return r; |
170 | return r; |
193 | } |
171 | } |
194 | for (i = 0; i < RADEON_IB_POOL_SIZE; i++) { |
172 | for (i = 0; i < RADEON_IB_POOL_SIZE; i++) { |
195 | unsigned offset; |
173 | unsigned offset; |
196 | 174 | ||
197 | offset = i * 64 * 1024; |
175 | offset = i * 64 * 1024; |
198 | rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset; |
176 | rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset; |
199 | rdev->ib_pool.ibs[i].ptr = ptr + offset; |
177 | rdev->ib_pool.ibs[i].ptr = ptr + offset; |
200 | rdev->ib_pool.ibs[i].idx = i; |
178 | rdev->ib_pool.ibs[i].idx = i; |
201 | rdev->ib_pool.ibs[i].length_dw = 0; |
179 | rdev->ib_pool.ibs[i].length_dw = 0; |
202 | INIT_LIST_HEAD(&rdev->ib_pool.ibs[i].list); |
180 | rdev->ib_pool.ibs[i].free = true; |
203 | } |
181 | } |
204 | bitmap_zero(rdev->ib_pool.alloc_bm, RADEON_IB_POOL_SIZE); |
182 | rdev->ib_pool.head_id = 0; |
205 | rdev->ib_pool.ready = true; |
183 | rdev->ib_pool.ready = true; |
206 | DRM_INFO("radeon: ib pool ready.\n"); |
184 | DRM_INFO("radeon: ib pool ready.\n"); |
207 | if (radeon_debugfs_ib_init(rdev)) { |
185 | if (radeon_debugfs_ib_init(rdev)) { |
208 | DRM_ERROR("Failed to register debugfs file for IB !\n"); |
186 | DRM_ERROR("Failed to register debugfs file for IB !\n"); |
209 | } |
187 | } |
210 | return r; |
188 | return r; |
211 | } |
189 | } |
212 | 190 | ||
213 | void radeon_ib_pool_fini(struct radeon_device *rdev) |
191 | void radeon_ib_pool_fini(struct radeon_device *rdev) |
214 | { |
192 | { |
215 | int r; |
193 | int r; |
216 | 194 | ||
217 | if (!rdev->ib_pool.ready) { |
195 | if (!rdev->ib_pool.ready) { |
218 | return; |
196 | return; |
219 | } |
197 | } |
220 | mutex_lock(&rdev->ib_pool.mutex); |
198 | mutex_lock(&rdev->ib_pool.mutex); |
221 | bitmap_zero(rdev->ib_pool.alloc_bm, RADEON_IB_POOL_SIZE); |
- | |
222 | if (rdev->ib_pool.robj) { |
199 | if (rdev->ib_pool.robj) { |
223 | r = radeon_bo_reserve(rdev->ib_pool.robj, false); |
200 | r = radeon_bo_reserve(rdev->ib_pool.robj, false); |
224 | if (likely(r == 0)) { |
201 | if (likely(r == 0)) { |
225 | radeon_bo_kunmap(rdev->ib_pool.robj); |
202 | radeon_bo_kunmap(rdev->ib_pool.robj); |
226 | radeon_bo_unpin(rdev->ib_pool.robj); |
203 | radeon_bo_unpin(rdev->ib_pool.robj); |
227 | radeon_bo_unreserve(rdev->ib_pool.robj); |
204 | radeon_bo_unreserve(rdev->ib_pool.robj); |
228 | } |
205 | } |
229 | radeon_bo_unref(&rdev->ib_pool.robj); |
206 | radeon_bo_unref(&rdev->ib_pool.robj); |
230 | rdev->ib_pool.robj = NULL; |
207 | rdev->ib_pool.robj = NULL; |
231 | } |
208 | } |
232 | mutex_unlock(&rdev->ib_pool.mutex); |
209 | mutex_unlock(&rdev->ib_pool.mutex); |
233 | } |
210 | } |
234 | 211 | ||
235 | 212 | ||
236 | /* |
213 | /* |
237 | * Ring. |
214 | * Ring. |
238 | */ |
215 | */ |
239 | void radeon_ring_free_size(struct radeon_device *rdev) |
216 | void radeon_ring_free_size(struct radeon_device *rdev) |
240 | { |
217 | { |
241 | if (rdev->family >= CHIP_R600) |
218 | if (rdev->family >= CHIP_R600) |
242 | rdev->cp.rptr = RREG32(R600_CP_RB_RPTR); |
219 | rdev->cp.rptr = RREG32(R600_CP_RB_RPTR); |
243 | else |
220 | else |
244 | rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR); |
221 | rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR); |
245 | /* This works because ring_size is a power of 2 */ |
222 | /* This works because ring_size is a power of 2 */ |
246 | rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4)); |
223 | rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4)); |
247 | rdev->cp.ring_free_dw -= rdev->cp.wptr; |
224 | rdev->cp.ring_free_dw -= rdev->cp.wptr; |
248 | rdev->cp.ring_free_dw &= rdev->cp.ptr_mask; |
225 | rdev->cp.ring_free_dw &= rdev->cp.ptr_mask; |
249 | if (!rdev->cp.ring_free_dw) { |
226 | if (!rdev->cp.ring_free_dw) { |
250 | rdev->cp.ring_free_dw = rdev->cp.ring_size / 4; |
227 | rdev->cp.ring_free_dw = rdev->cp.ring_size / 4; |
251 | } |
228 | } |
252 | } |
229 | } |
253 | 230 | ||
254 | int radeon_ring_lock(struct radeon_device *rdev, unsigned ndw) |
231 | int radeon_ring_lock(struct radeon_device *rdev, unsigned ndw) |
255 | { |
232 | { |
256 | int r; |
233 | int r; |
257 | 234 | ||
258 | /* Align requested size with padding so unlock_commit can |
235 | /* Align requested size with padding so unlock_commit can |
259 | * pad safely */ |
236 | * pad safely */ |
260 | ndw = (ndw + rdev->cp.align_mask) & ~rdev->cp.align_mask; |
237 | ndw = (ndw + rdev->cp.align_mask) & ~rdev->cp.align_mask; |
261 | mutex_lock(&rdev->cp.mutex); |
238 | mutex_lock(&rdev->cp.mutex); |
262 | while (ndw > (rdev->cp.ring_free_dw - 1)) { |
239 | while (ndw > (rdev->cp.ring_free_dw - 1)) { |
263 | radeon_ring_free_size(rdev); |
240 | radeon_ring_free_size(rdev); |
264 | if (ndw < rdev->cp.ring_free_dw) { |
241 | if (ndw < rdev->cp.ring_free_dw) { |
265 | break; |
242 | break; |
266 | } |
243 | } |
267 | // r = radeon_fence_wait_next(rdev); |
244 | // r = radeon_fence_wait_next(rdev); |
268 | // if (r) { |
245 | // if (r) { |
269 | // mutex_unlock(&rdev->cp.mutex); |
246 | // mutex_unlock(&rdev->cp.mutex); |
270 | // return r; |
247 | // return r; |
271 | // } |
248 | // } |
272 | } |
249 | } |
273 | rdev->cp.count_dw = ndw; |
250 | rdev->cp.count_dw = ndw; |
274 | rdev->cp.wptr_old = rdev->cp.wptr; |
251 | rdev->cp.wptr_old = rdev->cp.wptr; |
275 | return 0; |
252 | return 0; |
276 | } |
253 | } |
277 | 254 | ||
278 | void radeon_ring_unlock_commit(struct radeon_device *rdev) |
255 | void radeon_ring_unlock_commit(struct radeon_device *rdev) |
279 | { |
256 | { |
280 | unsigned count_dw_pad; |
257 | unsigned count_dw_pad; |
281 | unsigned i; |
258 | unsigned i; |
282 | 259 | ||
283 | /* We pad to match fetch size */ |
260 | /* We pad to match fetch size */ |
284 | count_dw_pad = (rdev->cp.align_mask + 1) - |
261 | count_dw_pad = (rdev->cp.align_mask + 1) - |
285 | (rdev->cp.wptr & rdev->cp.align_mask); |
262 | (rdev->cp.wptr & rdev->cp.align_mask); |
286 | for (i = 0; i < count_dw_pad; i++) { |
263 | for (i = 0; i < count_dw_pad; i++) { |
287 | radeon_ring_write(rdev, 2 << 30); |
264 | radeon_ring_write(rdev, 2 << 30); |
288 | } |
265 | } |
289 | DRM_MEMORYBARRIER(); |
266 | DRM_MEMORYBARRIER(); |
290 | radeon_cp_commit(rdev); |
267 | radeon_cp_commit(rdev); |
291 | mutex_unlock(&rdev->cp.mutex); |
268 | mutex_unlock(&rdev->cp.mutex); |
292 | } |
269 | } |
293 | 270 | ||
294 | void radeon_ring_unlock_undo(struct radeon_device *rdev) |
271 | void radeon_ring_unlock_undo(struct radeon_device *rdev) |
295 | { |
272 | { |
296 | rdev->cp.wptr = rdev->cp.wptr_old; |
273 | rdev->cp.wptr = rdev->cp.wptr_old; |
297 | mutex_unlock(&rdev->cp.mutex); |
274 | mutex_unlock(&rdev->cp.mutex); |
298 | } |
275 | } |
299 | 276 | ||
300 | int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size) |
277 | int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size) |
301 | { |
278 | { |
302 | int r; |
279 | int r; |
303 | 280 | ||
304 | ENTER(); |
281 | ENTER(); |
305 | 282 | ||
306 | rdev->cp.ring_size = ring_size; |
283 | rdev->cp.ring_size = ring_size; |
307 | /* Allocate ring buffer */ |
284 | /* Allocate ring buffer */ |
308 | if (rdev->cp.ring_obj == NULL) { |
285 | if (rdev->cp.ring_obj == NULL) { |
309 | r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true, |
286 | r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true, |
310 | RADEON_GEM_DOMAIN_GTT, |
287 | RADEON_GEM_DOMAIN_GTT, |
311 | &rdev->cp.ring_obj); |
288 | &rdev->cp.ring_obj); |
312 | if (r) { |
289 | if (r) { |
313 | dev_err(rdev->dev, "(%d) ring create failed\n", r); |
290 | dev_err(rdev->dev, "(%d) ring create failed\n", r); |
314 | return r; |
291 | return r; |
315 | } |
292 | } |
316 | r = radeon_bo_reserve(rdev->cp.ring_obj, false); |
293 | r = radeon_bo_reserve(rdev->cp.ring_obj, false); |
317 | if (unlikely(r != 0)) |
294 | if (unlikely(r != 0)) |
318 | return r; |
295 | return r; |
319 | r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT, |
296 | r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT, |
320 | &rdev->cp.gpu_addr); |
297 | &rdev->cp.gpu_addr); |
321 | if (r) { |
298 | if (r) { |
322 | radeon_bo_unreserve(rdev->cp.ring_obj); |
299 | radeon_bo_unreserve(rdev->cp.ring_obj); |
323 | dev_err(rdev->dev, "(%d) ring pin failed\n", r); |
300 | dev_err(rdev->dev, "(%d) ring pin failed\n", r); |
324 | return r; |
301 | return r; |
325 | } |
302 | } |
326 | r = radeon_bo_kmap(rdev->cp.ring_obj, |
303 | r = radeon_bo_kmap(rdev->cp.ring_obj, |
327 | (void **)&rdev->cp.ring); |
304 | (void **)&rdev->cp.ring); |
328 | radeon_bo_unreserve(rdev->cp.ring_obj); |
305 | radeon_bo_unreserve(rdev->cp.ring_obj); |
329 | if (r) { |
306 | if (r) { |
330 | dev_err(rdev->dev, "(%d) ring map failed\n", r); |
307 | dev_err(rdev->dev, "(%d) ring map failed\n", r); |
331 | return r; |
308 | return r; |
332 | } |
309 | } |
333 | } |
310 | } |
334 | rdev->cp.ptr_mask = (rdev->cp.ring_size / 4) - 1; |
311 | rdev->cp.ptr_mask = (rdev->cp.ring_size / 4) - 1; |
335 | rdev->cp.ring_free_dw = rdev->cp.ring_size / 4; |
312 | rdev->cp.ring_free_dw = rdev->cp.ring_size / 4; |
336 | 313 | ||
337 | LEAVE(); |
314 | LEAVE(); |
338 | 315 | ||
339 | return 0; |
316 | return 0; |
340 | } |
317 | } |
341 | 318 | ||
342 | void radeon_ring_fini(struct radeon_device *rdev) |
319 | void radeon_ring_fini(struct radeon_device *rdev) |
343 | { |
320 | { |
344 | int r; |
321 | int r; |
345 | 322 | ||
346 | mutex_lock(&rdev->cp.mutex); |
323 | mutex_lock(&rdev->cp.mutex); |
347 | if (rdev->cp.ring_obj) { |
324 | if (rdev->cp.ring_obj) { |
348 | r = radeon_bo_reserve(rdev->cp.ring_obj, false); |
325 | r = radeon_bo_reserve(rdev->cp.ring_obj, false); |
349 | if (likely(r == 0)) { |
326 | if (likely(r == 0)) { |
350 | radeon_bo_kunmap(rdev->cp.ring_obj); |
327 | radeon_bo_kunmap(rdev->cp.ring_obj); |
351 | radeon_bo_unpin(rdev->cp.ring_obj); |
328 | radeon_bo_unpin(rdev->cp.ring_obj); |
352 | radeon_bo_unreserve(rdev->cp.ring_obj); |
329 | radeon_bo_unreserve(rdev->cp.ring_obj); |
353 | } |
330 | } |
354 | radeon_bo_unref(&rdev->cp.ring_obj); |
331 | radeon_bo_unref(&rdev->cp.ring_obj); |
355 | rdev->cp.ring = NULL; |
332 | rdev->cp.ring = NULL; |
356 | rdev->cp.ring_obj = NULL; |
333 | rdev->cp.ring_obj = NULL; |
357 | } |
334 | } |
358 | mutex_unlock(&rdev->cp.mutex); |
335 | mutex_unlock(&rdev->cp.mutex); |
359 | } |
336 | } |
360 | 337 | ||
361 | 338 | ||
362 | /* |
339 | /* |
363 | * Debugfs info |
340 | * Debugfs info |
364 | */ |
341 | */ |
365 | #if defined(CONFIG_DEBUG_FS) |
342 | #if defined(CONFIG_DEBUG_FS) |
366 | static int radeon_debugfs_ib_info(struct seq_file *m, void *data) |
343 | static int radeon_debugfs_ib_info(struct seq_file *m, void *data) |
367 | { |
344 | { |
368 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
345 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
369 | struct radeon_ib *ib = node->info_ent->data; |
346 | struct radeon_ib *ib = node->info_ent->data; |
370 | unsigned i; |
347 | unsigned i; |
371 | 348 | ||
372 | if (ib == NULL) { |
349 | if (ib == NULL) { |
373 | return 0; |
350 | return 0; |
374 | } |
351 | } |
375 | seq_printf(m, "IB %04lu\n", ib->idx); |
352 | seq_printf(m, "IB %04u\n", ib->idx); |
376 | seq_printf(m, "IB fence %p\n", ib->fence); |
353 | seq_printf(m, "IB fence %p\n", ib->fence); |
377 | seq_printf(m, "IB size %05u dwords\n", ib->length_dw); |
354 | seq_printf(m, "IB size %05u dwords\n", ib->length_dw); |
378 | for (i = 0; i < ib->length_dw; i++) { |
355 | for (i = 0; i < ib->length_dw; i++) { |
379 | seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]); |
356 | seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]); |
380 | } |
357 | } |
381 | return 0; |
358 | return 0; |
382 | } |
359 | } |
383 | 360 | ||
384 | static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE]; |
361 | static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE]; |
385 | static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32]; |
362 | static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32]; |
386 | #endif |
363 | #endif |
387 | 364 | ||
388 | int radeon_debugfs_ib_init(struct radeon_device *rdev) |
365 | int radeon_debugfs_ib_init(struct radeon_device *rdev) |
389 | { |
366 | { |
390 | #if defined(CONFIG_DEBUG_FS) |
367 | #if defined(CONFIG_DEBUG_FS) |
391 | unsigned i; |
368 | unsigned i; |
392 | 369 | ||
393 | for (i = 0; i < RADEON_IB_POOL_SIZE; i++) { |
370 | for (i = 0; i < RADEON_IB_POOL_SIZE; i++) { |
394 | sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i); |
371 | sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i); |
395 | radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i]; |
372 | radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i]; |
396 | radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info; |
373 | radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info; |
397 | radeon_debugfs_ib_list[i].driver_features = 0; |
374 | radeon_debugfs_ib_list[i].driver_features = 0; |
398 | radeon_debugfs_ib_list[i].data = &rdev->ib_pool.ibs[i]; |
375 | radeon_debugfs_ib_list[i].data = &rdev->ib_pool.ibs[i]; |
399 | } |
376 | } |
400 | return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list, |
377 | return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list, |
401 | RADEON_IB_POOL_SIZE); |
378 | RADEON_IB_POOL_SIZE); |
402 | #else |
379 | #else |
403 | return 0; |
380 | return 0; |
404 | #endif |
381 | #endif |
405 | }>>><>>>>> |
382 | }>>><>>>>> |