drm/amdgpu: hdp flush&inval should always do
[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
4ff37a83 77 ib->vm_id = 0;
d38ceaf9
AD
78
79 return 0;
80}
81
82/**
83 * amdgpu_ib_free - free an IB (Indirect Buffer)
84 *
85 * @adev: amdgpu_device pointer
86 * @ib: IB object to free
cc55c45d 87 * @f: the fence SA bo need wait on for the ib alloation
d38ceaf9
AD
88 *
89 * Free an IB (all asics).
90 */
cc55c45d 91void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib, struct fence *f)
d38ceaf9 92{
cc55c45d 93 amdgpu_sa_bo_free(adev, &ib->sa_bo, f);
d38ceaf9
AD
94}
95
96/**
97 * amdgpu_ib_schedule - schedule an IB (Indirect Buffer) on the ring
98 *
99 * @adev: amdgpu_device pointer
100 * @num_ibs: number of IBs to schedule
101 * @ibs: IB objects to schedule
ec72b800 102 * @f: fence created during this submission
d38ceaf9
AD
103 *
104 * Schedule an IB on the associated ring (all asics).
105 * Returns 0 on success, error on failure.
106 *
107 * On SI, there are two parallel engines fed from the primary ring,
108 * the CE (Constant Engine) and the DE (Drawing Engine). Since
109 * resource descriptors have moved to memory, the CE allows you to
110 * prime the caches while the DE is updating register state so that
111 * the resource descriptors will be already in cache when the draw is
112 * processed. To accomplish this, the userspace driver submits two
113 * IBs, one for the CE and one for the DE. If there is a CE IB (called
114 * a CONST_IB), it will be put on the ring prior to the DE IB. Prior
115 * to SI there was just a DE IB.
116 */
b07c60c0 117int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
336d1f5e 118 struct amdgpu_ib *ibs, struct fence *last_vm_update,
c5637837 119 struct amdgpu_job *job, struct fence **f)
d38ceaf9 120{
b07c60c0 121 struct amdgpu_device *adev = ring->adev;
d38ceaf9 122 struct amdgpu_ib *ib = &ibs[0];
3cb485f3 123 struct amdgpu_ctx *ctx, *old_ctx;
73cfa5f5 124 struct fence *hwf;
c5637837 125 struct amdgpu_vm *vm = NULL;
03ccf481
ML
126 unsigned i, patch_offset = ~0;
127
d38ceaf9 128 int r = 0;
d38ceaf9
AD
129
130 if (num_ibs == 0)
131 return -EINVAL;
132
3cb485f3 133 ctx = ibs->ctx;
c5637837
ML
134 if (job) /* for domain0 job like ring test, ibs->job is not assigned */
135 vm = job->vm;
d919ad49 136
d38ceaf9
AD
137 if (!ring->ready) {
138 dev_err(adev->dev, "couldn't schedule ib\n");
139 return -EINVAL;
140 }
be86c606 141
4ff37a83 142 if (vm && !ibs->vm_id) {
8d0a7cea
CK
143 dev_err(adev->dev, "VM IB without ID\n");
144 return -EINVAL;
145 }
146
867d0517 147 r = amdgpu_ring_alloc(ring, 256 * num_ibs);
d38ceaf9
AD
148 if (r) {
149 dev_err(adev->dev, "scheduling IB failed (%d).\n", r);
150 return r;
151 }
152
03ccf481
ML
153 if (ring->type == AMDGPU_RING_TYPE_SDMA && ring->funcs->init_cond_exec)
154 patch_offset = amdgpu_ring_init_cond_exec(ring);
155
d38ceaf9
AD
156 if (vm) {
157 /* do context switch */
41d9eb2c
CK
158 r = amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr,
159 ib->gds_base, ib->gds_size,
160 ib->gws_base, ib->gws_size,
161 ib->oa_base, ib->oa_size);
162 if (r) {
163 amdgpu_ring_undo(ring);
164 return r;
165 }
e722b71a 166 }
d2edb07b 167
794ff571
ML
168 if (ring->funcs->emit_hdp_flush)
169 amdgpu_ring_emit_hdp_flush(ring);
170
128cff1a
ML
171 /* always set cond_exec_polling to CONTINUE */
172 *ring->cond_exe_cpu_addr = 1;
173
3cb485f3 174 old_ctx = ring->current_ctx;
d38ceaf9
AD
175 for (i = 0; i < num_ibs; ++i) {
176 ib = &ibs[i];
d38ceaf9 177 amdgpu_ring_emit_ib(ring, ib);
3cb485f3 178 ring->current_ctx = ctx;
d38ceaf9
AD
179 }
180
794ff571
ML
181 if (ring->funcs->emit_hdp_invalidate)
182 amdgpu_ring_emit_hdp_invalidate(ring);
11afbde8 183
73cfa5f5 184 r = amdgpu_fence_emit(ring, &hwf);
d38ceaf9
AD
185 if (r) {
186 dev_err(adev->dev, "failed to emit fence (%d)\n", r);
3cb485f3 187 ring->current_ctx = old_ctx;
971fe9a9
CK
188 if (ib->vm_id)
189 amdgpu_vm_reset_id(adev, ib->vm_id);
a27de35c 190 amdgpu_ring_undo(ring);
d38ceaf9
AD
191 return r;
192 }
193
194 /* wrap the last IB with fence */
195 if (ib->user) {
196 uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo);
197 addr += ib->user->offset;
5430a3ff 198 amdgpu_ring_emit_fence(ring, addr, ib->sequence,
890ee23f 199 AMDGPU_FENCE_FLAG_64BIT);
d38ceaf9
AD
200 }
201
ec72b800 202 if (f)
73cfa5f5 203 *f = fence_get(hwf);
ec72b800 204
03ccf481
ML
205 if (patch_offset != ~0 && ring->funcs->patch_cond_exec)
206 amdgpu_ring_patch_cond_exec(ring, patch_offset);
207
a27de35c 208 amdgpu_ring_commit(ring);
d38ceaf9
AD
209 return 0;
210}
211
212/**
213 * amdgpu_ib_pool_init - Init the IB (Indirect Buffer) pool
214 *
215 * @adev: amdgpu_device pointer
216 *
217 * Initialize the suballocator to manage a pool of memory
218 * for use as IBs (all asics).
219 * Returns 0 on success, error on failure.
220 */
221int amdgpu_ib_pool_init(struct amdgpu_device *adev)
222{
223 int r;
224
225 if (adev->ib_pool_ready) {
226 return 0;
227 }
228 r = amdgpu_sa_bo_manager_init(adev, &adev->ring_tmp_bo,
229 AMDGPU_IB_POOL_SIZE*64*1024,
230 AMDGPU_GPU_PAGE_SIZE,
231 AMDGPU_GEM_DOMAIN_GTT);
232 if (r) {
233 return r;
234 }
235
236 r = amdgpu_sa_bo_manager_start(adev, &adev->ring_tmp_bo);
237 if (r) {
238 return r;
239 }
240
241 adev->ib_pool_ready = true;
242 if (amdgpu_debugfs_sa_init(adev)) {
243 dev_err(adev->dev, "failed to register debugfs file for SA\n");
244 }
245 return 0;
246}
247
248/**
249 * amdgpu_ib_pool_fini - Free the IB (Indirect Buffer) pool
250 *
251 * @adev: amdgpu_device pointer
252 *
253 * Tear down the suballocator managing the pool of memory
254 * for use as IBs (all asics).
255 */
256void amdgpu_ib_pool_fini(struct amdgpu_device *adev)
257{
258 if (adev->ib_pool_ready) {
259 amdgpu_sa_bo_manager_suspend(adev, &adev->ring_tmp_bo);
260 amdgpu_sa_bo_manager_fini(adev, &adev->ring_tmp_bo);
261 adev->ib_pool_ready = false;
262 }
263}
264
265/**
266 * amdgpu_ib_ring_tests - test IBs on the rings
267 *
268 * @adev: amdgpu_device pointer
269 *
270 * Test an IB (Indirect Buffer) on each ring.
271 * If the test fails, disable the ring.
272 * Returns 0 on success, error if the primary GFX ring
273 * IB test fails.
274 */
275int amdgpu_ib_ring_tests(struct amdgpu_device *adev)
276{
277 unsigned i;
278 int r;
279
280 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
281 struct amdgpu_ring *ring = adev->rings[i];
282
283 if (!ring || !ring->ready)
284 continue;
285
286 r = amdgpu_ring_test_ib(ring);
287 if (r) {
288 ring->ready = false;
d38ceaf9
AD
289
290 if (ring == &adev->gfx.gfx_ring[0]) {
291 /* oh, oh, that's really bad */
292 DRM_ERROR("amdgpu: failed testing IB on GFX ring (%d).\n", r);
293 adev->accel_working = false;
294 return r;
295
296 } else {
297 /* still not good, but we can live with it */
298 DRM_ERROR("amdgpu: failed testing IB on ring %d (%d).\n", i, r);
299 }
300 }
301 }
302 return 0;
303}
304
305/*
306 * Debugfs info
307 */
308#if defined(CONFIG_DEBUG_FS)
309
310static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data)
311{
312 struct drm_info_node *node = (struct drm_info_node *) m->private;
313 struct drm_device *dev = node->minor->dev;
314 struct amdgpu_device *adev = dev->dev_private;
315
316 amdgpu_sa_bo_dump_debug_info(&adev->ring_tmp_bo, m);
317
318 return 0;
319
320}
321
06ab6832 322static const struct drm_info_list amdgpu_debugfs_sa_list[] = {
d38ceaf9
AD
323 {"amdgpu_sa_info", &amdgpu_debugfs_sa_info, 0, NULL},
324};
325
326#endif
327
328static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
329{
330#if defined(CONFIG_DEBUG_FS)
331 return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_sa_list, 1);
332#else
333 return 0;
334#endif
335}
This page took 0.134947 seconds and 5 git commands to generate.