2 * Copyright (C) 2015 Etnaviv Project
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 as published by
6 * the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program. If not, see <http://www.gnu.org/licenses/>.
17 #include <linux/dma-fence-array.h>
18 #include <linux/reservation.h>
19 #include <linux/sync_file.h>
20 #include "etnaviv_cmdbuf.h"
21 #include "etnaviv_drv.h"
22 #include "etnaviv_gpu.h"
23 #include "etnaviv_gem.h"
26 * Cmdstream submission:
29 #define BO_INVALID_FLAGS ~(ETNA_SUBMIT_BO_READ | ETNA_SUBMIT_BO_WRITE)
30 /* make sure these don't conflict w/ ETNAVIV_SUBMIT_BO_x */
31 #define BO_LOCKED 0x4000
32 #define BO_PINNED 0x2000
34 static struct etnaviv_gem_submit *submit_create(struct drm_device *dev,
35 struct etnaviv_gpu *gpu, size_t nr)
37 struct etnaviv_gem_submit *submit;
38 size_t sz = size_vstruct(nr, sizeof(submit->bos[0]), sizeof(*submit));
40 submit = kmalloc(sz, GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY);
45 /* initially, until copy_from_user() and bo lookup succeeds: */
49 ww_acquire_init(&submit->ticket, &reservation_ww_class);
55 static int submit_lookup_objects(struct etnaviv_gem_submit *submit,
56 struct drm_file *file, struct drm_etnaviv_gem_submit_bo *submit_bos,
59 struct drm_etnaviv_gem_submit_bo *bo;
63 spin_lock(&file->table_lock);
65 for (i = 0, bo = submit_bos; i < nr_bos; i++, bo++) {
66 struct drm_gem_object *obj;
68 if (bo->flags & BO_INVALID_FLAGS) {
69 DRM_ERROR("invalid flags: %x\n", bo->flags);
74 submit->bos[i].flags = bo->flags;
76 /* normally use drm_gem_object_lookup(), but for bulk lookup
77 * all under single table_lock just hit object_idr directly:
79 obj = idr_find(&file->object_idr, bo->handle);
81 DRM_ERROR("invalid handle %u at index %u\n",
88 * Take a refcount on the object. The file table lock
89 * prevents the object_idr's refcount on this being dropped.
91 drm_gem_object_reference(obj);
93 submit->bos[i].obj = to_etnaviv_bo(obj);
98 spin_unlock(&file->table_lock);
103 static void submit_unlock_object(struct etnaviv_gem_submit *submit, int i)
105 if (submit->bos[i].flags & BO_LOCKED) {
106 struct etnaviv_gem_object *etnaviv_obj = submit->bos[i].obj;
108 ww_mutex_unlock(&etnaviv_obj->resv->lock);
109 submit->bos[i].flags &= ~BO_LOCKED;
113 static int submit_lock_objects(struct etnaviv_gem_submit *submit)
115 int contended, slow_locked = -1, i, ret = 0;
118 for (i = 0; i < submit->nr_bos; i++) {
119 struct etnaviv_gem_object *etnaviv_obj = submit->bos[i].obj;
121 if (slow_locked == i)
126 if (!(submit->bos[i].flags & BO_LOCKED)) {
127 ret = ww_mutex_lock_interruptible(&etnaviv_obj->resv->lock,
129 if (ret == -EALREADY)
130 DRM_ERROR("BO at index %u already on submit list\n",
134 submit->bos[i].flags |= BO_LOCKED;
138 ww_acquire_done(&submit->ticket);
144 submit_unlock_object(submit, i);
147 submit_unlock_object(submit, slow_locked);
149 if (ret == -EDEADLK) {
150 struct etnaviv_gem_object *etnaviv_obj;
152 etnaviv_obj = submit->bos[contended].obj;
154 /* we lost out in a seqno race, lock and retry.. */
155 ret = ww_mutex_lock_slow_interruptible(&etnaviv_obj->resv->lock,
158 submit->bos[contended].flags |= BO_LOCKED;
159 slow_locked = contended;
167 static int submit_fence_sync(const struct etnaviv_gem_submit *submit)
169 unsigned int context = submit->gpu->fence_context;
172 for (i = 0; i < submit->nr_bos; i++) {
173 struct etnaviv_gem_object *etnaviv_obj = submit->bos[i].obj;
174 bool write = submit->bos[i].flags & ETNA_SUBMIT_BO_WRITE;
175 bool explicit = !(submit->flags & ETNA_SUBMIT_NO_IMPLICIT);
177 ret = etnaviv_gpu_fence_sync_obj(etnaviv_obj, context, write,
186 static void submit_unpin_objects(struct etnaviv_gem_submit *submit)
190 for (i = 0; i < submit->nr_bos; i++) {
191 if (submit->bos[i].flags & BO_PINNED)
192 etnaviv_gem_mapping_unreference(submit->bos[i].mapping);
194 submit->bos[i].mapping = NULL;
195 submit->bos[i].flags &= ~BO_PINNED;
199 static int submit_pin_objects(struct etnaviv_gem_submit *submit)
203 for (i = 0; i < submit->nr_bos; i++) {
204 struct etnaviv_gem_object *etnaviv_obj = submit->bos[i].obj;
205 struct etnaviv_vram_mapping *mapping;
207 mapping = etnaviv_gem_mapping_get(&etnaviv_obj->base,
209 if (IS_ERR(mapping)) {
210 ret = PTR_ERR(mapping);
214 submit->bos[i].flags |= BO_PINNED;
215 submit->bos[i].mapping = mapping;
221 static int submit_bo(struct etnaviv_gem_submit *submit, u32 idx,
222 struct etnaviv_gem_submit_bo **bo)
224 if (idx >= submit->nr_bos) {
225 DRM_ERROR("invalid buffer index: %u (out of %u)\n",
226 idx, submit->nr_bos);
230 *bo = &submit->bos[idx];
235 /* process the reloc's and patch up the cmdstream as needed: */
236 static int submit_reloc(struct etnaviv_gem_submit *submit, void *stream,
237 u32 size, const struct drm_etnaviv_gem_submit_reloc *relocs,
240 u32 i, last_offset = 0;
244 for (i = 0; i < nr_relocs; i++) {
245 const struct drm_etnaviv_gem_submit_reloc *r = relocs + i;
246 struct etnaviv_gem_submit_bo *bo;
249 if (unlikely(r->flags)) {
250 DRM_ERROR("invalid reloc flags\n");
254 if (r->submit_offset % 4) {
255 DRM_ERROR("non-aligned reloc offset: %u\n",
260 /* offset in dwords: */
261 off = r->submit_offset / 4;
263 if ((off >= size ) ||
264 (off < last_offset)) {
265 DRM_ERROR("invalid offset %u at reloc %u\n", off, i);
269 ret = submit_bo(submit, r->reloc_idx, &bo);
273 if (r->reloc_offset >= bo->obj->base.size - sizeof(*ptr)) {
274 DRM_ERROR("relocation %u outside object", i);
278 ptr[off] = bo->mapping->iova + r->reloc_offset;
286 static void submit_cleanup(struct etnaviv_gem_submit *submit)
290 for (i = 0; i < submit->nr_bos; i++) {
291 struct etnaviv_gem_object *etnaviv_obj = submit->bos[i].obj;
293 submit_unlock_object(submit, i);
294 drm_gem_object_unreference_unlocked(&etnaviv_obj->base);
297 ww_acquire_fini(&submit->ticket);
299 dma_fence_put(submit->fence);
303 int etnaviv_ioctl_gem_submit(struct drm_device *dev, void *data,
304 struct drm_file *file)
306 struct etnaviv_drm_private *priv = dev->dev_private;
307 struct drm_etnaviv_gem_submit *args = data;
308 struct drm_etnaviv_gem_submit_reloc *relocs;
309 struct drm_etnaviv_gem_submit_bo *bos;
310 struct etnaviv_gem_submit *submit;
311 struct etnaviv_cmdbuf *cmdbuf;
312 struct etnaviv_gpu *gpu;
313 struct dma_fence *in_fence = NULL;
314 struct sync_file *sync_file = NULL;
315 int out_fence_fd = -1;
319 if (args->pipe >= ETNA_MAX_PIPES)
322 gpu = priv->gpu[args->pipe];
326 if (args->stream_size % 4) {
327 DRM_ERROR("non-aligned cmdstream buffer size: %u\n",
332 if (args->exec_state != ETNA_PIPE_3D &&
333 args->exec_state != ETNA_PIPE_2D &&
334 args->exec_state != ETNA_PIPE_VG) {
335 DRM_ERROR("invalid exec_state: 0x%x\n", args->exec_state);
339 if (args->flags & ~ETNA_SUBMIT_FLAGS) {
340 DRM_ERROR("invalid flags: 0x%x\n", args->flags);
345 * Copy the command submission and bo array to kernel space in
346 * one go, and do this outside of any locks.
348 bos = drm_malloc_ab(args->nr_bos, sizeof(*bos));
349 relocs = drm_malloc_ab(args->nr_relocs, sizeof(*relocs));
350 stream = drm_malloc_ab(1, args->stream_size);
351 cmdbuf = etnaviv_cmdbuf_new(gpu->cmdbuf_suballoc,
352 ALIGN(args->stream_size, 8) + 8,
354 if (!bos || !relocs || !stream || !cmdbuf) {
356 goto err_submit_cmds;
359 cmdbuf->exec_state = args->exec_state;
360 cmdbuf->ctx = file->driver_priv;
362 ret = copy_from_user(bos, u64_to_user_ptr(args->bos),
363 args->nr_bos * sizeof(*bos));
366 goto err_submit_cmds;
369 ret = copy_from_user(relocs, u64_to_user_ptr(args->relocs),
370 args->nr_relocs * sizeof(*relocs));
373 goto err_submit_cmds;
376 ret = copy_from_user(stream, u64_to_user_ptr(args->stream),
380 goto err_submit_cmds;
383 if (args->flags & ETNA_SUBMIT_FENCE_FD_OUT) {
384 out_fence_fd = get_unused_fd_flags(O_CLOEXEC);
385 if (out_fence_fd < 0) {
387 goto err_submit_cmds;
391 submit = submit_create(dev, gpu, args->nr_bos);
394 goto err_submit_cmds;
397 submit->flags = args->flags;
399 ret = submit_lookup_objects(submit, file, bos, args->nr_bos);
401 goto err_submit_objects;
403 ret = submit_lock_objects(submit);
405 goto err_submit_objects;
407 if (!etnaviv_cmd_validate_one(gpu, stream, args->stream_size / 4,
408 relocs, args->nr_relocs)) {
410 goto err_submit_objects;
413 if (args->flags & ETNA_SUBMIT_FENCE_FD_IN) {
414 in_fence = sync_file_get_fence(args->fence_fd);
417 goto err_submit_objects;
421 * Wait if the fence is from a foreign context, or if the fence
422 * array contains any fence from a foreign context.
424 if (!dma_fence_match_context(in_fence, gpu->fence_context)) {
425 ret = dma_fence_wait(in_fence, true);
427 goto err_submit_objects;
431 ret = submit_fence_sync(submit);
433 goto err_submit_objects;
435 ret = submit_pin_objects(submit);
439 ret = submit_reloc(submit, stream, args->stream_size / 4,
440 relocs, args->nr_relocs);
444 memcpy(cmdbuf->vaddr, stream, args->stream_size);
445 cmdbuf->user_size = ALIGN(args->stream_size, 8);
447 ret = etnaviv_gpu_submit(gpu, submit, cmdbuf);
451 if (args->flags & ETNA_SUBMIT_FENCE_FD_OUT) {
453 * This can be improved: ideally we want to allocate the sync
454 * file before kicking off the GPU job and just attach the
455 * fence to the sync file here, eliminating the ENOMEM
456 * possibility at this stage.
458 sync_file = sync_file_create(submit->fence);
463 fd_install(out_fence_fd, sync_file->file);
466 args->fence_fd = out_fence_fd;
467 args->fence = submit->fence->seqno;
470 submit_unpin_objects(submit);
473 * If we're returning -EAGAIN, it may be due to the userptr code
474 * wanting to run its workqueue outside of any locks. Flush our
475 * workqueue to ensure that it is run in a timely manner.
478 flush_workqueue(priv->wq);
482 dma_fence_put(in_fence);
483 submit_cleanup(submit);
486 if (ret && (out_fence_fd >= 0))
487 put_unused_fd(out_fence_fd);
488 /* if we still own the cmdbuf */
490 etnaviv_cmdbuf_free(cmdbuf);
492 drm_free_large(stream);
496 drm_free_large(relocs);