Merge drm-fixes into drm-next.
[deliverable/linux.git] / drivers / gpu / drm / amd / amdgpu / amdgpu_ib.c
CommitLineData
d38ceaf9
AD
1/*
2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
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
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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,
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
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
22 * OTHER DEALINGS IN THE SOFTWARE.
23 *
24 * Authors: Dave Airlie
25 * Alex Deucher
26 * Jerome Glisse
27 * Christian König
28 */
29#include <linux/seq_file.h>
30#include <linux/slab.h>
31#include <drm/drmP.h>
32#include <drm/amdgpu_drm.h>
33#include "amdgpu.h"
34#include "atom.h"
35
36/*
37 * IB
38 * IBs (Indirect Buffers) and areas of GPU accessible memory where
39 * commands are stored. You can put a pointer to the IB in the
40 * command ring and the hw will fetch the commands from the IB
41 * and execute them. Generally userspace acceleration drivers
42 * produce command buffers which are send to the kernel and
43 * put in IBs for execution by the requested ring.
44 */
45static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
46
47/**
48 * amdgpu_ib_get - request an IB (Indirect Buffer)
49 *
50 * @ring: ring index the IB is associated with
51 * @size: requested IB size
52 * @ib: IB object returned
53 *
54 * Request an IB (all asics). IBs are allocated using the
55 * suballocator.
56 * Returns 0 on success, error on failure.
57 */
b07c60c0 58int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
d38ceaf9
AD
59 unsigned size, struct amdgpu_ib *ib)
60{
d38ceaf9
AD
61 int r;
62
63 if (size) {
bbf0b345 64 r = amdgpu_sa_bo_new(&adev->ring_tmp_bo,
d38ceaf9
AD
65 &ib->sa_bo, size, 256);
66 if (r) {
67 dev_err(adev->dev, "failed to get a new IB (%d)\n", r);
68 return r;
69 }
70
71 ib->ptr = amdgpu_sa_bo_cpu_addr(ib->sa_bo);
72
73 if (!vm)
74 ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
d38ceaf9
AD
75 }
76
d38ceaf9 77 ib->vm = vm;
4ff37a83 78 ib->vm_id = 0;
d38ceaf9
AD
79
80 return 0;
81}
82
83/**
84 * amdgpu_ib_free - free an IB (Indirect Buffer)
85 *
86 * @adev: amdgpu_device pointer
87 * @ib: IB object to free
88 *
89 * Free an IB (all asics).
90 */
91void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib)
92{
4ce9891e 93 amdgpu_sa_bo_free(adev, &ib->sa_bo, &ib->fence->base);
6ef68c17
CK
94 if (ib->fence)
95 fence_put(&ib->fence->base);
d38ceaf9
AD
96}
97
98/**
99 * amdgpu_ib_schedule - schedule an IB (Indirect Buffer) on the ring
100 *
101 * @adev: amdgpu_device pointer
102 * @num_ibs: number of IBs to schedule
103 * @ibs: IB objects to schedule
104 * @owner: owner for creating the fences
ec72b800 105 * @f: fence created during this submission
d38ceaf9
AD
106 *
107 * Schedule an IB on the associated ring (all asics).
108 * Returns 0 on success, error on failure.
109 *
110 * On SI, there are two parallel engines fed from the primary ring,
111 * the CE (Constant Engine) and the DE (Drawing Engine). Since
112 * resource descriptors have moved to memory, the CE allows you to
113 * prime the caches while the DE is updating register state so that
114 * the resource descriptors will be already in cache when the draw is
115 * processed. To accomplish this, the userspace driver submits two
116 * IBs, one for the CE and one for the DE. If there is a CE IB (called
117 * a CONST_IB), it will be put on the ring prior to the DE IB. Prior
118 * to SI there was just a DE IB.
119 */
b07c60c0 120int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
ec72b800 121 struct amdgpu_ib *ibs, void *owner,
e86f9cee 122 struct fence *last_vm_update,
ec72b800 123 struct fence **f)
d38ceaf9 124{
b07c60c0 125 struct amdgpu_device *adev = ring->adev;
d38ceaf9 126 struct amdgpu_ib *ib = &ibs[0];
3cb485f3 127 struct amdgpu_ctx *ctx, *old_ctx;
d919ad49 128 struct amdgpu_vm *vm;
d38ceaf9
AD
129 unsigned i;
130 int r = 0;
d38ceaf9
AD
131
132 if (num_ibs == 0)
133 return -EINVAL;
134
3cb485f3 135 ctx = ibs->ctx;
d919ad49
CK
136 vm = ibs->vm;
137
d38ceaf9
AD
138 if (!ring->ready) {
139 dev_err(adev->dev, "couldn't schedule ib\n");
140 return -EINVAL;
141 }
be86c606 142
4ff37a83 143 if (vm && !ibs->vm_id) {
8d0a7cea
CK
144 dev_err(adev->dev, "VM IB without ID\n");
145 return -EINVAL;
146 }
147
867d0517 148 r = amdgpu_ring_alloc(ring, 256 * num_ibs);
d38ceaf9
AD
149 if (r) {
150 dev_err(adev->dev, "scheduling IB failed (%d).\n", r);
151 return r;
152 }
153
d38ceaf9
AD
154 if (vm) {
155 /* do context switch */
4ff37a83 156 amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr);
d38ceaf9 157
e722b71a 158 if (ring->funcs->emit_gds_switch)
4ff37a83 159 amdgpu_ring_emit_gds_switch(ring, ib->vm_id,
e722b71a 160 ib->gds_base, ib->gds_size,
161 ib->gws_base, ib->gws_size,
162 ib->oa_base, ib->oa_size);
d38ceaf9 163
e722b71a 164 if (ring->funcs->emit_hdp_flush)
165 amdgpu_ring_emit_hdp_flush(ring);
166 }
d2edb07b 167
3cb485f3 168 old_ctx = ring->current_ctx;
d38ceaf9
AD
169 for (i = 0; i < num_ibs; ++i) {
170 ib = &ibs[i];
171
b07c60c0 172 if (ib->ctx != ctx || ib->vm != vm) {
3cb485f3 173 ring->current_ctx = old_ctx;
a27de35c 174 amdgpu_ring_undo(ring);
d38ceaf9
AD
175 return -EINVAL;
176 }
d38ceaf9 177 amdgpu_ring_emit_ib(ring, ib);
3cb485f3 178 ring->current_ctx = ctx;
d38ceaf9
AD
179 }
180
181 r = amdgpu_fence_emit(ring, owner, &ib->fence);
182 if (r) {
183 dev_err(adev->dev, "failed to emit fence (%d)\n", r);
3cb485f3 184 ring->current_ctx = old_ctx;
a27de35c 185 amdgpu_ring_undo(ring);
d38ceaf9
AD
186 return r;
187 }
188
189 /* wrap the last IB with fence */
190 if (ib->user) {
191 uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo);
192 addr += ib->user->offset;
5430a3ff 193 amdgpu_ring_emit_fence(ring, addr, ib->sequence,
890ee23f 194 AMDGPU_FENCE_FLAG_64BIT);
d38ceaf9
AD
195 }
196
ec72b800
CK
197 if (f)
198 *f = fence_get(&ib->fence->base);
199
a27de35c 200 amdgpu_ring_commit(ring);
d38ceaf9
AD
201 return 0;
202}
203
204/**
205 * amdgpu_ib_pool_init - Init the IB (Indirect Buffer) pool
206 *
207 * @adev: amdgpu_device pointer
208 *
209 * Initialize the suballocator to manage a pool of memory
210 * for use as IBs (all asics).
211 * Returns 0 on success, error on failure.
212 */
213int amdgpu_ib_pool_init(struct amdgpu_device *adev)
214{
215 int r;
216
217 if (adev->ib_pool_ready) {
218 return 0;
219 }
220 r = amdgpu_sa_bo_manager_init(adev, &adev->ring_tmp_bo,
221 AMDGPU_IB_POOL_SIZE*64*1024,
222 AMDGPU_GPU_PAGE_SIZE,
223 AMDGPU_GEM_DOMAIN_GTT);
224 if (r) {
225 return r;
226 }
227
228 r = amdgpu_sa_bo_manager_start(adev, &adev->ring_tmp_bo);
229 if (r) {
230 return r;
231 }
232
233 adev->ib_pool_ready = true;
234 if (amdgpu_debugfs_sa_init(adev)) {
235 dev_err(adev->dev, "failed to register debugfs file for SA\n");
236 }
237 return 0;
238}
239
240/**
241 * amdgpu_ib_pool_fini - Free the IB (Indirect Buffer) pool
242 *
243 * @adev: amdgpu_device pointer
244 *
245 * Tear down the suballocator managing the pool of memory
246 * for use as IBs (all asics).
247 */
248void amdgpu_ib_pool_fini(struct amdgpu_device *adev)
249{
250 if (adev->ib_pool_ready) {
251 amdgpu_sa_bo_manager_suspend(adev, &adev->ring_tmp_bo);
252 amdgpu_sa_bo_manager_fini(adev, &adev->ring_tmp_bo);
253 adev->ib_pool_ready = false;
254 }
255}
256
257/**
258 * amdgpu_ib_ring_tests - test IBs on the rings
259 *
260 * @adev: amdgpu_device pointer
261 *
262 * Test an IB (Indirect Buffer) on each ring.
263 * If the test fails, disable the ring.
264 * Returns 0 on success, error if the primary GFX ring
265 * IB test fails.
266 */
267int amdgpu_ib_ring_tests(struct amdgpu_device *adev)
268{
269 unsigned i;
270 int r;
271
272 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
273 struct amdgpu_ring *ring = adev->rings[i];
274
275 if (!ring || !ring->ready)
276 continue;
277
278 r = amdgpu_ring_test_ib(ring);
279 if (r) {
280 ring->ready = false;
d38ceaf9
AD
281
282 if (ring == &adev->gfx.gfx_ring[0]) {
283 /* oh, oh, that's really bad */
284 DRM_ERROR("amdgpu: failed testing IB on GFX ring (%d).\n", r);
285 adev->accel_working = false;
286 return r;
287
288 } else {
289 /* still not good, but we can live with it */
290 DRM_ERROR("amdgpu: failed testing IB on ring %d (%d).\n", i, r);
291 }
292 }
293 }
294 return 0;
295}
296
297/*
298 * Debugfs info
299 */
300#if defined(CONFIG_DEBUG_FS)
301
302static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data)
303{
304 struct drm_info_node *node = (struct drm_info_node *) m->private;
305 struct drm_device *dev = node->minor->dev;
306 struct amdgpu_device *adev = dev->dev_private;
307
308 amdgpu_sa_bo_dump_debug_info(&adev->ring_tmp_bo, m);
309
310 return 0;
311
312}
313
314static struct drm_info_list amdgpu_debugfs_sa_list[] = {
315 {"amdgpu_sa_info", &amdgpu_debugfs_sa_info, 0, NULL},
316};
317
318#endif
319
320static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
321{
322#if defined(CONFIG_DEBUG_FS)
323 return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_sa_list, 1);
324#else
325 return 0;
326#endif
327}
This page took 0.092555 seconds and 5 git commands to generate.