2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 * Authors: Dave Airlie
28 #include <linux/seq_file.h>
30 #include "radeon_drm.h"
31 #include "radeon_reg.h"
35 int radeon_debugfs_ib_init(struct radeon_device *rdev);
37 void radeon_ib_bogus_cleanup(struct radeon_device *rdev)
39 struct radeon_ib *ib, *n;
41 list_for_each_entry_safe(ib, n, &rdev->ib_pool.bogus_ib, list) {
48 void radeon_ib_bogus_add(struct radeon_device *rdev, struct radeon_ib *ib)
50 struct radeon_ib *bib;
52 bib = kmalloc(sizeof(*bib), GFP_KERNEL);
55 bib->ptr = vmalloc(ib->length_dw * 4);
56 if (bib->ptr == NULL) {
60 memcpy(bib->ptr, ib->ptr, ib->length_dw * 4);
61 bib->length_dw = ib->length_dw;
62 mutex_lock(&rdev->ib_pool.mutex);
63 list_add_tail(&bib->list, &rdev->ib_pool.bogus_ib);
64 mutex_unlock(&rdev->ib_pool.mutex);
70 int radeon_ib_get(struct radeon_device *rdev, struct radeon_ib **ib)
72 struct radeon_fence *fence;
73 struct radeon_ib *nib;
77 r = radeon_fence_create(rdev, &fence);
79 dev_err(rdev->dev, "failed to create fence for new IB\n");
82 mutex_lock(&rdev->ib_pool.mutex);
83 for (i = rdev->ib_pool.head_id, c = 0, nib = NULL; c < RADEON_IB_POOL_SIZE; c++, i++) {
84 i &= (RADEON_IB_POOL_SIZE - 1);
85 if (rdev->ib_pool.ibs[i].free) {
86 nib = &rdev->ib_pool.ibs[i];
91 /* This should never happen, it means we allocated all
92 * IB and haven't scheduled one yet, return EBUSY to
93 * userspace hoping that on ioctl recall we get better
96 dev_err(rdev->dev, "no free indirect buffer !\n");
97 mutex_unlock(&rdev->ib_pool.mutex);
98 radeon_fence_unref(&fence);
101 rdev->ib_pool.head_id = (nib->idx + 1) & (RADEON_IB_POOL_SIZE - 1);
104 mutex_unlock(&rdev->ib_pool.mutex);
105 r = radeon_fence_wait(nib->fence, false);
107 dev_err(rdev->dev, "error waiting fence of IB(%u:0x%016lX:%u)\n",
108 nib->idx, (unsigned long)nib->gpu_addr, nib->length_dw);
109 mutex_lock(&rdev->ib_pool.mutex);
111 mutex_unlock(&rdev->ib_pool.mutex);
112 radeon_fence_unref(&fence);
115 mutex_lock(&rdev->ib_pool.mutex);
117 radeon_fence_unref(&nib->fence);
120 mutex_unlock(&rdev->ib_pool.mutex);
125 void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib)
127 struct radeon_ib *tmp = *ib;
133 if (!tmp->fence->emited)
134 radeon_fence_unref(&tmp->fence);
135 mutex_lock(&rdev->ib_pool.mutex);
137 mutex_unlock(&rdev->ib_pool.mutex);
140 int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
144 if (!ib->length_dw || !rdev->cp.ready) {
145 /* TODO: Nothings in the ib we should report. */
146 DRM_ERROR("radeon: couldn't schedule IB(%u).\n", ib->idx);
150 /* 64 dwords should be enough for fence too */
151 r = radeon_ring_lock(rdev, 64);
153 DRM_ERROR("radeon: scheduling IB failled (%d).\n", r);
156 radeon_ring_ib_execute(rdev, ib);
157 radeon_fence_emit(rdev, ib->fence);
158 mutex_lock(&rdev->ib_pool.mutex);
159 /* once scheduled IB is considered free and protected by the fence */
161 mutex_unlock(&rdev->ib_pool.mutex);
162 radeon_ring_unlock_commit(rdev);
166 int radeon_ib_pool_init(struct radeon_device *rdev)
173 if (rdev->ib_pool.robj)
175 INIT_LIST_HEAD(&rdev->ib_pool.bogus_ib);
176 /* Allocate 1M object buffer */
177 r = radeon_bo_create(rdev, NULL, RADEON_IB_POOL_SIZE*64*1024,
178 true, RADEON_GEM_DOMAIN_GTT,
179 &rdev->ib_pool.robj);
181 DRM_ERROR("radeon: failed to ib pool (%d).\n", r);
184 r = radeon_bo_reserve(rdev->ib_pool.robj, false);
185 if (unlikely(r != 0))
187 r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr);
189 radeon_bo_unreserve(rdev->ib_pool.robj);
190 DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r);
193 r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr);
194 radeon_bo_unreserve(rdev->ib_pool.robj);
196 DRM_ERROR("radeon: failed to map ib poll (%d).\n", r);
199 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
202 offset = i * 64 * 1024;
203 rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset;
204 rdev->ib_pool.ibs[i].ptr = ptr + offset;
205 rdev->ib_pool.ibs[i].idx = i;
206 rdev->ib_pool.ibs[i].length_dw = 0;
207 rdev->ib_pool.ibs[i].free = true;
209 rdev->ib_pool.head_id = 0;
210 rdev->ib_pool.ready = true;
211 DRM_INFO("radeon: ib pool ready.\n");
212 if (radeon_debugfs_ib_init(rdev)) {
213 DRM_ERROR("Failed to register debugfs file for IB !\n");
218 void radeon_ib_pool_fini(struct radeon_device *rdev)
222 if (!rdev->ib_pool.ready) {
225 mutex_lock(&rdev->ib_pool.mutex);
226 radeon_ib_bogus_cleanup(rdev);
228 if (rdev->ib_pool.robj) {
229 r = radeon_bo_reserve(rdev->ib_pool.robj, false);
230 if (likely(r == 0)) {
231 radeon_bo_kunmap(rdev->ib_pool.robj);
232 radeon_bo_unpin(rdev->ib_pool.robj);
233 radeon_bo_unreserve(rdev->ib_pool.robj);
235 radeon_bo_unref(&rdev->ib_pool.robj);
236 rdev->ib_pool.robj = NULL;
238 mutex_unlock(&rdev->ib_pool.mutex);
245 void radeon_ring_free_size(struct radeon_device *rdev)
247 if (rdev->family >= CHIP_R600)
248 rdev->cp.rptr = RREG32(R600_CP_RB_RPTR);
250 rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
251 /* This works because ring_size is a power of 2 */
252 rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4));
253 rdev->cp.ring_free_dw -= rdev->cp.wptr;
254 rdev->cp.ring_free_dw &= rdev->cp.ptr_mask;
255 if (!rdev->cp.ring_free_dw) {
256 rdev->cp.ring_free_dw = rdev->cp.ring_size / 4;
260 int radeon_ring_lock(struct radeon_device *rdev, unsigned ndw)
264 /* Align requested size with padding so unlock_commit can
266 ndw = (ndw + rdev->cp.align_mask) & ~rdev->cp.align_mask;
267 mutex_lock(&rdev->cp.mutex);
268 while (ndw > (rdev->cp.ring_free_dw - 1)) {
269 radeon_ring_free_size(rdev);
270 if (ndw < rdev->cp.ring_free_dw) {
273 r = radeon_fence_wait_next(rdev);
275 mutex_unlock(&rdev->cp.mutex);
279 rdev->cp.count_dw = ndw;
280 rdev->cp.wptr_old = rdev->cp.wptr;
284 void radeon_ring_unlock_commit(struct radeon_device *rdev)
286 unsigned count_dw_pad;
289 /* We pad to match fetch size */
290 count_dw_pad = (rdev->cp.align_mask + 1) -
291 (rdev->cp.wptr & rdev->cp.align_mask);
292 for (i = 0; i < count_dw_pad; i++) {
293 radeon_ring_write(rdev, 2 << 30);
296 radeon_cp_commit(rdev);
297 mutex_unlock(&rdev->cp.mutex);
300 void radeon_ring_unlock_undo(struct radeon_device *rdev)
302 rdev->cp.wptr = rdev->cp.wptr_old;
303 mutex_unlock(&rdev->cp.mutex);
306 int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size)
310 rdev->cp.ring_size = ring_size;
311 /* Allocate ring buffer */
312 if (rdev->cp.ring_obj == NULL) {
313 r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true,
314 RADEON_GEM_DOMAIN_GTT,
317 dev_err(rdev->dev, "(%d) ring create failed\n", r);
320 r = radeon_bo_reserve(rdev->cp.ring_obj, false);
321 if (unlikely(r != 0))
323 r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT,
326 radeon_bo_unreserve(rdev->cp.ring_obj);
327 dev_err(rdev->dev, "(%d) ring pin failed\n", r);
330 r = radeon_bo_kmap(rdev->cp.ring_obj,
331 (void **)&rdev->cp.ring);
332 radeon_bo_unreserve(rdev->cp.ring_obj);
334 dev_err(rdev->dev, "(%d) ring map failed\n", r);
338 rdev->cp.ptr_mask = (rdev->cp.ring_size / 4) - 1;
339 rdev->cp.ring_free_dw = rdev->cp.ring_size / 4;
343 void radeon_ring_fini(struct radeon_device *rdev)
347 mutex_lock(&rdev->cp.mutex);
348 if (rdev->cp.ring_obj) {
349 r = radeon_bo_reserve(rdev->cp.ring_obj, false);
350 if (likely(r == 0)) {
351 radeon_bo_kunmap(rdev->cp.ring_obj);
352 radeon_bo_unpin(rdev->cp.ring_obj);
353 radeon_bo_unreserve(rdev->cp.ring_obj);
355 radeon_bo_unref(&rdev->cp.ring_obj);
356 rdev->cp.ring = NULL;
357 rdev->cp.ring_obj = NULL;
359 mutex_unlock(&rdev->cp.mutex);
366 #if defined(CONFIG_DEBUG_FS)
367 static int radeon_debugfs_ib_info(struct seq_file *m, void *data)
369 struct drm_info_node *node = (struct drm_info_node *) m->private;
370 struct radeon_ib *ib = node->info_ent->data;
376 seq_printf(m, "IB %04u\n", ib->idx);
377 seq_printf(m, "IB fence %p\n", ib->fence);
378 seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
379 for (i = 0; i < ib->length_dw; i++) {
380 seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
385 static int radeon_debugfs_ib_bogus_info(struct seq_file *m, void *data)
387 struct drm_info_node *node = (struct drm_info_node *) m->private;
388 struct radeon_device *rdev = node->info_ent->data;
389 struct radeon_ib *ib;
392 mutex_lock(&rdev->ib_pool.mutex);
393 if (list_empty(&rdev->ib_pool.bogus_ib)) {
394 mutex_unlock(&rdev->ib_pool.mutex);
395 seq_printf(m, "no bogus IB recorded\n");
398 ib = list_first_entry(&rdev->ib_pool.bogus_ib, struct radeon_ib, list);
399 list_del_init(&ib->list);
400 mutex_unlock(&rdev->ib_pool.mutex);
401 seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
402 for (i = 0; i < ib->length_dw; i++) {
403 seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
410 static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE];
411 static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32];
413 static struct drm_info_list radeon_debugfs_ib_bogus_info_list[] = {
414 {"radeon_ib_bogus", radeon_debugfs_ib_bogus_info, 0, NULL},
418 int radeon_debugfs_ib_init(struct radeon_device *rdev)
420 #if defined(CONFIG_DEBUG_FS)
424 radeon_debugfs_ib_bogus_info_list[0].data = rdev;
425 r = radeon_debugfs_add_files(rdev, radeon_debugfs_ib_bogus_info_list, 1);
428 for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
429 sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i);
430 radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i];
431 radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info;
432 radeon_debugfs_ib_list[i].driver_features = 0;
433 radeon_debugfs_ib_list[i].data = &rdev->ib_pool.ibs[i];
435 return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list,
436 RADEON_IB_POOL_SIZE);