1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2012 Avionic Design GmbH
4 * Copyright (C) 2012-2016 NVIDIA CORPORATION. All rights reserved.
7 #include <linux/aperture.h>
8 #include <linux/bitops.h>
9 #include <linux/host1x.h>
10 #include <linux/idr.h>
11 #include <linux/iommu.h>
12 #include <linux/module.h>
13 #include <linux/platform_device.h>
14 #include <linux/pm_runtime.h>
16 #include <drm/clients/drm_client_setup.h>
17 #include <drm/drm_atomic.h>
18 #include <drm/drm_atomic_helper.h>
19 #include <drm/drm_debugfs.h>
20 #include <drm/drm_drv.h>
21 #include <drm/drm_fourcc.h>
22 #include <drm/drm_framebuffer.h>
23 #include <drm/drm_ioctl.h>
24 #include <drm/drm_prime.h>
25 #include <drm/drm_vblank.h>
27 #if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
28 #include <asm/dma-iommu.h>
36 #define DRIVER_NAME "tegra"
37 #define DRIVER_DESC "NVIDIA Tegra graphics"
38 #define DRIVER_MAJOR 1
39 #define DRIVER_MINOR 0
40 #define DRIVER_PATCHLEVEL 0
42 #define CARVEOUT_SZ SZ_64M
43 #define CDMA_GATHER_FETCHES_MAX_NB 16383
45 static int tegra_atomic_check(struct drm_device *drm,
46 struct drm_atomic_state *state)
50 err = drm_atomic_helper_check(drm, state);
54 return tegra_display_hub_atomic_check(drm, state);
57 static const struct drm_mode_config_funcs tegra_drm_mode_config_funcs = {
58 .fb_create = tegra_fb_create,
59 .atomic_check = tegra_atomic_check,
60 .atomic_commit = drm_atomic_helper_commit,
63 static void tegra_atomic_post_commit(struct drm_device *drm,
64 struct drm_atomic_state *old_state)
66 struct drm_crtc_state *old_crtc_state __maybe_unused;
67 struct drm_crtc *crtc;
70 for_each_old_crtc_in_state(old_state, crtc, old_crtc_state, i)
71 tegra_crtc_atomic_post_commit(crtc, old_state);
74 static void tegra_atomic_commit_tail(struct drm_atomic_state *old_state)
76 struct drm_device *drm = old_state->dev;
77 struct tegra_drm *tegra = drm->dev_private;
80 bool fence_cookie = dma_fence_begin_signalling();
82 drm_atomic_helper_commit_modeset_disables(drm, old_state);
83 tegra_display_hub_atomic_commit(drm, old_state);
84 drm_atomic_helper_commit_planes(drm, old_state, 0);
85 drm_atomic_helper_commit_modeset_enables(drm, old_state);
86 drm_atomic_helper_commit_hw_done(old_state);
87 dma_fence_end_signalling(fence_cookie);
88 drm_atomic_helper_wait_for_vblanks(drm, old_state);
89 drm_atomic_helper_cleanup_planes(drm, old_state);
91 drm_atomic_helper_commit_tail_rpm(old_state);
94 tegra_atomic_post_commit(drm, old_state);
97 static const struct drm_mode_config_helper_funcs
98 tegra_drm_mode_config_helpers = {
99 .atomic_commit_tail = tegra_atomic_commit_tail,
102 static int tegra_drm_open(struct drm_device *drm, struct drm_file *filp)
104 struct tegra_drm_file *fpriv;
106 fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL);
110 idr_init_base(&fpriv->legacy_contexts, 1);
111 xa_init_flags(&fpriv->contexts, XA_FLAGS_ALLOC1);
112 xa_init(&fpriv->syncpoints);
113 mutex_init(&fpriv->lock);
114 filp->driver_priv = fpriv;
119 static void tegra_drm_context_free(struct tegra_drm_context *context)
121 context->client->ops->close_channel(context);
122 pm_runtime_put(context->client->base.dev);
126 static int host1x_reloc_copy_from_user(struct host1x_reloc *dest,
127 struct drm_tegra_reloc __user *src,
128 struct drm_device *drm,
129 struct drm_file *file)
134 err = get_user(cmdbuf, &src->cmdbuf.handle);
138 err = get_user(dest->cmdbuf.offset, &src->cmdbuf.offset);
142 err = get_user(target, &src->target.handle);
146 err = get_user(dest->target.offset, &src->target.offset);
150 err = get_user(dest->shift, &src->shift);
154 dest->flags = HOST1X_RELOC_READ | HOST1X_RELOC_WRITE;
156 dest->cmdbuf.bo = tegra_gem_lookup(file, cmdbuf);
157 if (!dest->cmdbuf.bo)
160 dest->target.bo = tegra_gem_lookup(file, target);
161 if (!dest->target.bo)
167 int tegra_drm_submit(struct tegra_drm_context *context,
168 struct drm_tegra_submit *args, struct drm_device *drm,
169 struct drm_file *file)
171 struct host1x_client *client = &context->client->base;
172 unsigned int num_cmdbufs = args->num_cmdbufs;
173 unsigned int num_relocs = args->num_relocs;
174 struct drm_tegra_cmdbuf __user *user_cmdbufs;
175 struct drm_tegra_reloc __user *user_relocs;
176 struct drm_tegra_syncpt __user *user_syncpt;
177 struct drm_tegra_syncpt syncpt;
178 struct host1x *host1x = dev_get_drvdata(drm->dev->parent);
179 struct drm_gem_object **refs;
180 struct host1x_syncpt *sp = NULL;
181 struct host1x_job *job;
182 unsigned int num_refs;
185 user_cmdbufs = u64_to_user_ptr(args->cmdbufs);
186 user_relocs = u64_to_user_ptr(args->relocs);
187 user_syncpt = u64_to_user_ptr(args->syncpts);
189 /* We don't yet support other than one syncpt_incr struct per submit */
190 if (args->num_syncpts != 1)
193 /* We don't yet support waitchks */
194 if (args->num_waitchks != 0)
197 job = host1x_job_alloc(context->channel, args->num_cmdbufs,
198 args->num_relocs, false);
202 job->num_relocs = args->num_relocs;
203 job->client = client;
204 job->class = client->class;
205 job->serialize = true;
206 job->syncpt_recovery = true;
209 * Track referenced BOs so that they can be unreferenced after the
210 * submission is complete.
212 num_refs = num_cmdbufs + num_relocs * 2;
214 refs = kmalloc_array(num_refs, sizeof(*refs), GFP_KERNEL);
220 /* reuse as an iterator later */
223 while (num_cmdbufs) {
224 struct drm_tegra_cmdbuf cmdbuf;
225 struct host1x_bo *bo;
226 struct tegra_bo *obj;
229 if (copy_from_user(&cmdbuf, user_cmdbufs, sizeof(cmdbuf))) {
235 * The maximum number of CDMA gather fetches is 16383, a higher
236 * value means the words count is malformed.
238 if (cmdbuf.words > CDMA_GATHER_FETCHES_MAX_NB) {
243 bo = tegra_gem_lookup(file, cmdbuf.handle);
249 offset = (u64)cmdbuf.offset + (u64)cmdbuf.words * sizeof(u32);
250 obj = host1x_to_tegra_bo(bo);
251 refs[num_refs++] = &obj->gem;
254 * Gather buffer base address must be 4-bytes aligned,
255 * unaligned offset is malformed and cause commands stream
256 * corruption on the buffer address relocation.
258 if (offset & 3 || offset > obj->gem.size) {
263 host1x_job_add_gather(job, bo, cmdbuf.words, cmdbuf.offset);
268 /* copy and resolve relocations from submit */
269 while (num_relocs--) {
270 struct host1x_reloc *reloc;
271 struct tegra_bo *obj;
273 err = host1x_reloc_copy_from_user(&job->relocs[num_relocs],
274 &user_relocs[num_relocs], drm,
279 reloc = &job->relocs[num_relocs];
280 obj = host1x_to_tegra_bo(reloc->cmdbuf.bo);
281 refs[num_refs++] = &obj->gem;
284 * The unaligned cmdbuf offset will cause an unaligned write
285 * during of the relocations patching, corrupting the commands
288 if (reloc->cmdbuf.offset & 3 ||
289 reloc->cmdbuf.offset >= obj->gem.size) {
294 obj = host1x_to_tegra_bo(reloc->target.bo);
295 refs[num_refs++] = &obj->gem;
297 if (reloc->target.offset >= obj->gem.size) {
303 if (copy_from_user(&syncpt, user_syncpt, sizeof(syncpt))) {
308 /* Syncpoint ref will be dropped on job release. */
309 sp = host1x_syncpt_get_by_id(host1x, syncpt.id);
315 job->is_addr_reg = context->client->ops->is_addr_reg;
316 job->is_valid_class = context->client->ops->is_valid_class;
317 job->syncpt_incrs = syncpt.incrs;
319 job->timeout = 10000;
321 if (args->timeout && args->timeout < 10000)
322 job->timeout = args->timeout;
324 err = host1x_job_pin(job, context->client->base.dev);
328 err = host1x_job_submit(job);
330 host1x_job_unpin(job);
334 args->fence = job->syncpt_end;
338 drm_gem_object_put(refs[num_refs]);
348 #ifdef CONFIG_DRM_TEGRA_STAGING
349 static int tegra_gem_create(struct drm_device *drm, void *data,
350 struct drm_file *file)
352 struct drm_tegra_gem_create *args = data;
355 bo = tegra_bo_create_with_handle(file, drm, args->size, args->flags,
363 static int tegra_gem_mmap(struct drm_device *drm, void *data,
364 struct drm_file *file)
366 struct drm_tegra_gem_mmap *args = data;
367 struct drm_gem_object *gem;
370 gem = drm_gem_object_lookup(file, args->handle);
374 bo = to_tegra_bo(gem);
376 args->offset = drm_vma_node_offset_addr(&bo->gem.vma_node);
378 drm_gem_object_put(gem);
383 static int tegra_syncpt_read(struct drm_device *drm, void *data,
384 struct drm_file *file)
386 struct host1x *host = dev_get_drvdata(drm->dev->parent);
387 struct drm_tegra_syncpt_read *args = data;
388 struct host1x_syncpt *sp;
390 sp = host1x_syncpt_get_by_id_noref(host, args->id);
394 args->value = host1x_syncpt_read_min(sp);
398 static int tegra_syncpt_incr(struct drm_device *drm, void *data,
399 struct drm_file *file)
401 struct host1x *host1x = dev_get_drvdata(drm->dev->parent);
402 struct drm_tegra_syncpt_incr *args = data;
403 struct host1x_syncpt *sp;
405 sp = host1x_syncpt_get_by_id_noref(host1x, args->id);
409 return host1x_syncpt_incr(sp);
412 static int tegra_syncpt_wait(struct drm_device *drm, void *data,
413 struct drm_file *file)
415 struct host1x *host1x = dev_get_drvdata(drm->dev->parent);
416 struct drm_tegra_syncpt_wait *args = data;
417 struct host1x_syncpt *sp;
419 sp = host1x_syncpt_get_by_id_noref(host1x, args->id);
423 return host1x_syncpt_wait(sp, args->thresh,
424 msecs_to_jiffies(args->timeout),
428 static int tegra_client_open(struct tegra_drm_file *fpriv,
429 struct tegra_drm_client *client,
430 struct tegra_drm_context *context)
434 err = pm_runtime_resume_and_get(client->base.dev);
438 err = client->ops->open_channel(client, context);
440 pm_runtime_put(client->base.dev);
444 err = idr_alloc(&fpriv->legacy_contexts, context, 1, 0, GFP_KERNEL);
446 client->ops->close_channel(context);
447 pm_runtime_put(client->base.dev);
451 context->client = client;
457 static int tegra_open_channel(struct drm_device *drm, void *data,
458 struct drm_file *file)
460 struct tegra_drm_file *fpriv = file->driver_priv;
461 struct tegra_drm *tegra = drm->dev_private;
462 struct drm_tegra_open_channel *args = data;
463 struct tegra_drm_context *context;
464 struct tegra_drm_client *client;
467 context = kzalloc(sizeof(*context), GFP_KERNEL);
471 mutex_lock(&fpriv->lock);
473 list_for_each_entry(client, &tegra->clients, list)
474 if (client->base.class == args->client) {
475 err = tegra_client_open(fpriv, client, context);
479 args->context = context->id;
486 mutex_unlock(&fpriv->lock);
490 static int tegra_close_channel(struct drm_device *drm, void *data,
491 struct drm_file *file)
493 struct tegra_drm_file *fpriv = file->driver_priv;
494 struct drm_tegra_close_channel *args = data;
495 struct tegra_drm_context *context;
498 mutex_lock(&fpriv->lock);
500 context = idr_find(&fpriv->legacy_contexts, args->context);
506 idr_remove(&fpriv->legacy_contexts, context->id);
507 tegra_drm_context_free(context);
510 mutex_unlock(&fpriv->lock);
514 static int tegra_get_syncpt(struct drm_device *drm, void *data,
515 struct drm_file *file)
517 struct tegra_drm_file *fpriv = file->driver_priv;
518 struct drm_tegra_get_syncpt *args = data;
519 struct tegra_drm_context *context;
520 struct host1x_syncpt *syncpt;
523 mutex_lock(&fpriv->lock);
525 context = idr_find(&fpriv->legacy_contexts, args->context);
531 if (args->index >= context->client->base.num_syncpts) {
536 syncpt = context->client->base.syncpts[args->index];
537 args->id = host1x_syncpt_id(syncpt);
540 mutex_unlock(&fpriv->lock);
544 static int tegra_submit(struct drm_device *drm, void *data,
545 struct drm_file *file)
547 struct tegra_drm_file *fpriv = file->driver_priv;
548 struct drm_tegra_submit *args = data;
549 struct tegra_drm_context *context;
552 mutex_lock(&fpriv->lock);
554 context = idr_find(&fpriv->legacy_contexts, args->context);
560 err = context->client->ops->submit(context, args, drm, file);
563 mutex_unlock(&fpriv->lock);
567 static int tegra_get_syncpt_base(struct drm_device *drm, void *data,
568 struct drm_file *file)
570 struct tegra_drm_file *fpriv = file->driver_priv;
571 struct drm_tegra_get_syncpt_base *args = data;
572 struct tegra_drm_context *context;
573 struct host1x_syncpt_base *base;
574 struct host1x_syncpt *syncpt;
577 mutex_lock(&fpriv->lock);
579 context = idr_find(&fpriv->legacy_contexts, args->context);
585 if (args->syncpt >= context->client->base.num_syncpts) {
590 syncpt = context->client->base.syncpts[args->syncpt];
592 base = host1x_syncpt_get_base(syncpt);
598 args->id = host1x_syncpt_base_id(base);
601 mutex_unlock(&fpriv->lock);
605 static int tegra_gem_set_tiling(struct drm_device *drm, void *data,
606 struct drm_file *file)
608 struct drm_tegra_gem_set_tiling *args = data;
609 enum tegra_bo_tiling_mode mode;
610 struct drm_gem_object *gem;
611 unsigned long value = 0;
614 switch (args->mode) {
615 case DRM_TEGRA_GEM_TILING_MODE_PITCH:
616 mode = TEGRA_BO_TILING_MODE_PITCH;
618 if (args->value != 0)
623 case DRM_TEGRA_GEM_TILING_MODE_TILED:
624 mode = TEGRA_BO_TILING_MODE_TILED;
626 if (args->value != 0)
631 case DRM_TEGRA_GEM_TILING_MODE_BLOCK:
632 mode = TEGRA_BO_TILING_MODE_BLOCK;
644 gem = drm_gem_object_lookup(file, args->handle);
648 bo = to_tegra_bo(gem);
650 bo->tiling.mode = mode;
651 bo->tiling.value = value;
653 drm_gem_object_put(gem);
658 static int tegra_gem_get_tiling(struct drm_device *drm, void *data,
659 struct drm_file *file)
661 struct drm_tegra_gem_get_tiling *args = data;
662 struct drm_gem_object *gem;
666 gem = drm_gem_object_lookup(file, args->handle);
670 bo = to_tegra_bo(gem);
672 switch (bo->tiling.mode) {
673 case TEGRA_BO_TILING_MODE_PITCH:
674 args->mode = DRM_TEGRA_GEM_TILING_MODE_PITCH;
678 case TEGRA_BO_TILING_MODE_TILED:
679 args->mode = DRM_TEGRA_GEM_TILING_MODE_TILED;
683 case TEGRA_BO_TILING_MODE_BLOCK:
684 args->mode = DRM_TEGRA_GEM_TILING_MODE_BLOCK;
685 args->value = bo->tiling.value;
693 drm_gem_object_put(gem);
698 static int tegra_gem_set_flags(struct drm_device *drm, void *data,
699 struct drm_file *file)
701 struct drm_tegra_gem_set_flags *args = data;
702 struct drm_gem_object *gem;
705 if (args->flags & ~DRM_TEGRA_GEM_FLAGS)
708 gem = drm_gem_object_lookup(file, args->handle);
712 bo = to_tegra_bo(gem);
715 if (args->flags & DRM_TEGRA_GEM_BOTTOM_UP)
716 bo->flags |= TEGRA_BO_BOTTOM_UP;
718 drm_gem_object_put(gem);
723 static int tegra_gem_get_flags(struct drm_device *drm, void *data,
724 struct drm_file *file)
726 struct drm_tegra_gem_get_flags *args = data;
727 struct drm_gem_object *gem;
730 gem = drm_gem_object_lookup(file, args->handle);
734 bo = to_tegra_bo(gem);
737 if (bo->flags & TEGRA_BO_BOTTOM_UP)
738 args->flags |= DRM_TEGRA_GEM_BOTTOM_UP;
740 drm_gem_object_put(gem);
746 static const struct drm_ioctl_desc tegra_drm_ioctls[] = {
747 #ifdef CONFIG_DRM_TEGRA_STAGING
748 DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_OPEN, tegra_drm_ioctl_channel_open,
750 DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_CLOSE, tegra_drm_ioctl_channel_close,
752 DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_MAP, tegra_drm_ioctl_channel_map,
754 DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_UNMAP, tegra_drm_ioctl_channel_unmap,
756 DRM_IOCTL_DEF_DRV(TEGRA_CHANNEL_SUBMIT, tegra_drm_ioctl_channel_submit,
758 DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_ALLOCATE, tegra_drm_ioctl_syncpoint_allocate,
760 DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_FREE, tegra_drm_ioctl_syncpoint_free,
762 DRM_IOCTL_DEF_DRV(TEGRA_SYNCPOINT_WAIT, tegra_drm_ioctl_syncpoint_wait,
765 DRM_IOCTL_DEF_DRV(TEGRA_GEM_CREATE, tegra_gem_create, DRM_RENDER_ALLOW),
766 DRM_IOCTL_DEF_DRV(TEGRA_GEM_MMAP, tegra_gem_mmap, DRM_RENDER_ALLOW),
767 DRM_IOCTL_DEF_DRV(TEGRA_SYNCPT_READ, tegra_syncpt_read,
769 DRM_IOCTL_DEF_DRV(TEGRA_SYNCPT_INCR, tegra_syncpt_incr,
771 DRM_IOCTL_DEF_DRV(TEGRA_SYNCPT_WAIT, tegra_syncpt_wait,
773 DRM_IOCTL_DEF_DRV(TEGRA_OPEN_CHANNEL, tegra_open_channel,
775 DRM_IOCTL_DEF_DRV(TEGRA_CLOSE_CHANNEL, tegra_close_channel,
777 DRM_IOCTL_DEF_DRV(TEGRA_GET_SYNCPT, tegra_get_syncpt,
779 DRM_IOCTL_DEF_DRV(TEGRA_SUBMIT, tegra_submit,
781 DRM_IOCTL_DEF_DRV(TEGRA_GET_SYNCPT_BASE, tegra_get_syncpt_base,
783 DRM_IOCTL_DEF_DRV(TEGRA_GEM_SET_TILING, tegra_gem_set_tiling,
785 DRM_IOCTL_DEF_DRV(TEGRA_GEM_GET_TILING, tegra_gem_get_tiling,
787 DRM_IOCTL_DEF_DRV(TEGRA_GEM_SET_FLAGS, tegra_gem_set_flags,
789 DRM_IOCTL_DEF_DRV(TEGRA_GEM_GET_FLAGS, tegra_gem_get_flags,
794 static const struct file_operations tegra_drm_fops = {
795 .owner = THIS_MODULE,
797 .release = drm_release,
798 .unlocked_ioctl = drm_ioctl,
799 .mmap = tegra_drm_mmap,
802 .compat_ioctl = drm_compat_ioctl,
803 .llseek = noop_llseek,
804 .fop_flags = FOP_UNSIGNED_OFFSET,
807 static int tegra_drm_context_cleanup(int id, void *p, void *data)
809 struct tegra_drm_context *context = p;
811 tegra_drm_context_free(context);
816 static void tegra_drm_postclose(struct drm_device *drm, struct drm_file *file)
818 struct tegra_drm_file *fpriv = file->driver_priv;
820 mutex_lock(&fpriv->lock);
821 idr_for_each(&fpriv->legacy_contexts, tegra_drm_context_cleanup, NULL);
822 tegra_drm_uapi_close_file(fpriv);
823 mutex_unlock(&fpriv->lock);
825 idr_destroy(&fpriv->legacy_contexts);
826 mutex_destroy(&fpriv->lock);
830 #ifdef CONFIG_DEBUG_FS
831 static int tegra_debugfs_framebuffers(struct seq_file *s, void *data)
833 struct drm_info_node *node = (struct drm_info_node *)s->private;
834 struct drm_device *drm = node->minor->dev;
835 struct drm_framebuffer *fb;
837 mutex_lock(&drm->mode_config.fb_lock);
839 list_for_each_entry(fb, &drm->mode_config.fb_list, head) {
840 seq_printf(s, "%3d: user size: %d x %d, depth %d, %d bpp, refcount %d\n",
841 fb->base.id, fb->width, fb->height,
843 fb->format->cpp[0] * 8,
844 drm_framebuffer_read_refcount(fb));
847 mutex_unlock(&drm->mode_config.fb_lock);
852 static int tegra_debugfs_iova(struct seq_file *s, void *data)
854 struct drm_info_node *node = (struct drm_info_node *)s->private;
855 struct drm_device *drm = node->minor->dev;
856 struct tegra_drm *tegra = drm->dev_private;
857 struct drm_printer p = drm_seq_file_printer(s);
860 mutex_lock(&tegra->mm_lock);
861 drm_mm_print(&tegra->mm, &p);
862 mutex_unlock(&tegra->mm_lock);
868 static struct drm_info_list tegra_debugfs_list[] = {
869 { "framebuffers", tegra_debugfs_framebuffers, 0 },
870 { "iova", tegra_debugfs_iova, 0 },
873 static void tegra_debugfs_init(struct drm_minor *minor)
875 drm_debugfs_create_files(tegra_debugfs_list,
876 ARRAY_SIZE(tegra_debugfs_list),
877 minor->debugfs_root, minor);
881 static const struct drm_driver tegra_drm_driver = {
882 .driver_features = DRIVER_MODESET | DRIVER_GEM |
883 DRIVER_ATOMIC | DRIVER_RENDER | DRIVER_SYNCOBJ,
884 .open = tegra_drm_open,
885 .postclose = tegra_drm_postclose,
887 #if defined(CONFIG_DEBUG_FS)
888 .debugfs_init = tegra_debugfs_init,
891 .gem_prime_import = tegra_gem_prime_import,
893 .dumb_create = tegra_bo_dumb_create,
895 TEGRA_FBDEV_DRIVER_OPS,
897 .ioctls = tegra_drm_ioctls,
898 .num_ioctls = ARRAY_SIZE(tegra_drm_ioctls),
899 .fops = &tegra_drm_fops,
903 .major = DRIVER_MAJOR,
904 .minor = DRIVER_MINOR,
905 .patchlevel = DRIVER_PATCHLEVEL,
908 int tegra_drm_register_client(struct tegra_drm *tegra,
909 struct tegra_drm_client *client)
912 * When MLOCKs are implemented, change to allocate a shared channel
913 * only when MLOCKs are disabled.
915 client->shared_channel = host1x_channel_request(&client->base);
916 if (!client->shared_channel)
919 mutex_lock(&tegra->clients_lock);
920 list_add_tail(&client->list, &tegra->clients);
922 mutex_unlock(&tegra->clients_lock);
927 int tegra_drm_unregister_client(struct tegra_drm *tegra,
928 struct tegra_drm_client *client)
930 mutex_lock(&tegra->clients_lock);
931 list_del_init(&client->list);
933 mutex_unlock(&tegra->clients_lock);
935 if (client->shared_channel)
936 host1x_channel_put(client->shared_channel);
941 int host1x_client_iommu_attach(struct host1x_client *client)
943 struct iommu_domain *domain = iommu_get_domain_for_dev(client->dev);
944 struct drm_device *drm = dev_get_drvdata(client->host);
945 struct tegra_drm *tegra = drm->dev_private;
946 struct iommu_group *group = NULL;
949 #if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
950 if (client->dev->archdata.mapping) {
951 struct dma_iommu_mapping *mapping =
952 to_dma_iommu_mapping(client->dev);
953 arm_iommu_detach_device(client->dev);
954 arm_iommu_release_mapping(mapping);
956 domain = iommu_get_domain_for_dev(client->dev);
961 * If the host1x client is already attached to an IOMMU domain that is
962 * not the shared IOMMU domain, don't try to attach it to a different
963 * domain. This allows using the IOMMU-backed DMA API.
965 if (domain && domain->type != IOMMU_DOMAIN_IDENTITY &&
966 domain != tegra->domain)
970 group = iommu_group_get(client->dev);
974 if (domain != tegra->domain) {
975 err = iommu_attach_group(tegra->domain, group);
977 iommu_group_put(group);
982 tegra->use_explicit_iommu = true;
985 client->group = group;
990 void host1x_client_iommu_detach(struct host1x_client *client)
992 struct drm_device *drm = dev_get_drvdata(client->host);
993 struct tegra_drm *tegra = drm->dev_private;
994 struct iommu_domain *domain;
998 * Devices that are part of the same group may no longer be
999 * attached to a domain at this point because their group may
1000 * have been detached by an earlier client.
1002 domain = iommu_get_domain_for_dev(client->dev);
1004 iommu_detach_group(tegra->domain, client->group);
1006 iommu_group_put(client->group);
1007 client->group = NULL;
1011 void *tegra_drm_alloc(struct tegra_drm *tegra, size_t size, dma_addr_t *dma)
1019 size = iova_align(&tegra->carveout.domain, size);
1021 size = PAGE_ALIGN(size);
1023 gfp = GFP_KERNEL | __GFP_ZERO;
1024 if (!tegra->domain) {
1026 * Many units only support 32-bit addresses, even on 64-bit
1027 * SoCs. If there is no IOMMU to translate into a 32-bit IO
1028 * virtual address space, force allocations to be in the
1029 * lower 32-bit range.
1034 virt = (void *)__get_free_pages(gfp, get_order(size));
1036 return ERR_PTR(-ENOMEM);
1038 if (!tegra->domain) {
1040 * If IOMMU is disabled, devices address physical memory
1043 *dma = virt_to_phys(virt);
1047 alloc = alloc_iova(&tegra->carveout.domain,
1048 size >> tegra->carveout.shift,
1049 tegra->carveout.limit, true);
1055 *dma = iova_dma_addr(&tegra->carveout.domain, alloc);
1056 err = iommu_map(tegra->domain, *dma, virt_to_phys(virt),
1057 size, IOMMU_READ | IOMMU_WRITE, GFP_KERNEL);
1064 __free_iova(&tegra->carveout.domain, alloc);
1066 free_pages((unsigned long)virt, get_order(size));
1068 return ERR_PTR(err);
1071 void tegra_drm_free(struct tegra_drm *tegra, size_t size, void *virt,
1075 size = iova_align(&tegra->carveout.domain, size);
1077 size = PAGE_ALIGN(size);
1079 if (tegra->domain) {
1080 iommu_unmap(tegra->domain, dma, size);
1081 free_iova(&tegra->carveout.domain,
1082 iova_pfn(&tegra->carveout.domain, dma));
1085 free_pages((unsigned long)virt, get_order(size));
1088 static bool host1x_drm_wants_iommu(struct host1x_device *dev)
1090 struct host1x *host1x = dev_get_drvdata(dev->dev.parent);
1091 struct iommu_domain *domain;
1093 /* Our IOMMU usage policy doesn't currently play well with GART */
1094 if (of_machine_is_compatible("nvidia,tegra20"))
1098 * If the Tegra DRM clients are backed by an IOMMU, push buffers are
1099 * likely to be allocated beyond the 32-bit boundary if sufficient
1100 * system memory is available. This is problematic on earlier Tegra
1101 * generations where host1x supports a maximum of 32 address bits in
1102 * the GATHER opcode. In this case, unless host1x is behind an IOMMU
1103 * as well it won't be able to process buffers allocated beyond the
1106 * The DMA API will use bounce buffers in this case, so that could
1107 * perhaps still be made to work, even if less efficient, but there
1108 * is another catch: in order to perform cache maintenance on pages
1109 * allocated for discontiguous buffers we need to map and unmap the
1110 * SG table representing these buffers. This is fine for something
1111 * small like a push buffer, but it exhausts the bounce buffer pool
1112 * (typically on the order of a few MiB) for framebuffers (many MiB
1113 * for any modern resolution).
1115 * Work around this by making sure that Tegra DRM clients only use
1116 * an IOMMU if the parent host1x also uses an IOMMU.
1118 * Note that there's still a small gap here that we don't cover: if
1119 * the DMA API is backed by an IOMMU there's no way to control which
1120 * device is attached to an IOMMU and which isn't, except via wiring
1121 * up the device tree appropriately. This is considered an problem
1122 * of integration, so care must be taken for the DT to be consistent.
1124 domain = iommu_get_domain_for_dev(dev->dev.parent);
1127 * Tegra20 and Tegra30 don't support addressing memory beyond the
1128 * 32-bit boundary, so the regular GATHER opcodes will always be
1129 * sufficient and whether or not the host1x is attached to an IOMMU
1132 if (!domain && host1x_get_dma_mask(host1x) <= DMA_BIT_MASK(32))
1135 return domain != NULL;
1138 static int host1x_drm_probe(struct host1x_device *dev)
1140 struct device *dma_dev = dev->dev.parent;
1141 struct tegra_drm *tegra;
1142 struct drm_device *drm;
1145 drm = drm_dev_alloc(&tegra_drm_driver, &dev->dev);
1147 return PTR_ERR(drm);
1149 tegra = kzalloc(sizeof(*tegra), GFP_KERNEL);
1155 if (host1x_drm_wants_iommu(dev) && device_iommu_mapped(dma_dev)) {
1156 tegra->domain = iommu_paging_domain_alloc(dma_dev);
1157 if (IS_ERR(tegra->domain)) {
1158 err = PTR_ERR(tegra->domain);
1162 err = iova_cache_get();
1167 mutex_init(&tegra->clients_lock);
1168 INIT_LIST_HEAD(&tegra->clients);
1170 dev_set_drvdata(&dev->dev, drm);
1171 drm->dev_private = tegra;
1174 drm_mode_config_init(drm);
1176 drm->mode_config.min_width = 0;
1177 drm->mode_config.min_height = 0;
1178 drm->mode_config.max_width = 0;
1179 drm->mode_config.max_height = 0;
1181 drm->mode_config.normalize_zpos = true;
1183 drm->mode_config.funcs = &tegra_drm_mode_config_funcs;
1184 drm->mode_config.helper_private = &tegra_drm_mode_config_helpers;
1186 drm_kms_helper_poll_init(drm);
1188 err = host1x_device_init(dev);
1193 * Now that all display controller have been initialized, the maximum
1194 * supported resolution is known and the bitmask for horizontal and
1195 * vertical bitfields can be computed.
1197 tegra->hmask = drm->mode_config.max_width - 1;
1198 tegra->vmask = drm->mode_config.max_height - 1;
1200 if (tegra->use_explicit_iommu) {
1201 u64 carveout_start, carveout_end, gem_start, gem_end;
1202 u64 dma_mask = dma_get_mask(&dev->dev);
1203 dma_addr_t start, end;
1204 unsigned long order;
1206 start = tegra->domain->geometry.aperture_start & dma_mask;
1207 end = tegra->domain->geometry.aperture_end & dma_mask;
1210 gem_end = end - CARVEOUT_SZ;
1211 carveout_start = gem_end + 1;
1214 order = __ffs(tegra->domain->pgsize_bitmap);
1215 init_iova_domain(&tegra->carveout.domain, 1UL << order,
1216 carveout_start >> order);
1218 tegra->carveout.shift = iova_shift(&tegra->carveout.domain);
1219 tegra->carveout.limit = carveout_end >> tegra->carveout.shift;
1221 drm_mm_init(&tegra->mm, gem_start, gem_end - gem_start + 1);
1222 mutex_init(&tegra->mm_lock);
1224 DRM_DEBUG_DRIVER("IOMMU apertures:\n");
1225 DRM_DEBUG_DRIVER(" GEM: %#llx-%#llx\n", gem_start, gem_end);
1226 DRM_DEBUG_DRIVER(" Carveout: %#llx-%#llx\n", carveout_start,
1228 } else if (tegra->domain) {
1229 iommu_domain_free(tegra->domain);
1230 tegra->domain = NULL;
1235 err = tegra_display_hub_prepare(tegra->hub);
1240 /* syncpoints are used for full 32-bit hardware VBLANK counters */
1241 drm->max_vblank_count = 0xffffffff;
1243 err = drm_vblank_init(drm, drm->mode_config.num_crtc);
1247 drm_mode_config_reset(drm);
1250 * Only take over from a potential firmware framebuffer if any CRTCs
1251 * have been registered. This must not be a fatal error because there
1252 * are other accelerators that are exposed via this driver.
1254 * Another case where this happens is on Tegra234 where the display
1255 * hardware is no longer part of the host1x complex, so this driver
1256 * will not expose any modesetting features.
1258 if (drm->mode_config.num_crtc > 0) {
1259 err = aperture_remove_all_conflicting_devices(tegra_drm_driver.name);
1264 * Indicate to userspace that this doesn't expose any display
1267 drm->driver_features &= ~(DRIVER_MODESET | DRIVER_ATOMIC);
1270 err = drm_dev_register(drm, 0);
1274 drm_client_setup(drm, NULL);
1280 tegra_display_hub_cleanup(tegra->hub);
1282 if (tegra->domain) {
1283 mutex_destroy(&tegra->mm_lock);
1284 drm_mm_takedown(&tegra->mm);
1285 put_iova_domain(&tegra->carveout.domain);
1289 host1x_device_exit(dev);
1291 drm_kms_helper_poll_fini(drm);
1292 drm_mode_config_cleanup(drm);
1295 iommu_domain_free(tegra->domain);
1303 static int host1x_drm_remove(struct host1x_device *dev)
1305 struct drm_device *drm = dev_get_drvdata(&dev->dev);
1306 struct tegra_drm *tegra = drm->dev_private;
1309 drm_dev_unregister(drm);
1311 drm_kms_helper_poll_fini(drm);
1312 drm_atomic_helper_shutdown(drm);
1313 drm_mode_config_cleanup(drm);
1316 tegra_display_hub_cleanup(tegra->hub);
1318 err = host1x_device_exit(dev);
1320 dev_err(&dev->dev, "host1x device cleanup failed: %d\n", err);
1322 if (tegra->domain) {
1323 mutex_destroy(&tegra->mm_lock);
1324 drm_mm_takedown(&tegra->mm);
1325 put_iova_domain(&tegra->carveout.domain);
1327 iommu_domain_free(tegra->domain);
1336 static void host1x_drm_shutdown(struct host1x_device *dev)
1338 drm_atomic_helper_shutdown(dev_get_drvdata(&dev->dev));
1341 #ifdef CONFIG_PM_SLEEP
1342 static int host1x_drm_suspend(struct device *dev)
1344 struct drm_device *drm = dev_get_drvdata(dev);
1346 return drm_mode_config_helper_suspend(drm);
1349 static int host1x_drm_resume(struct device *dev)
1351 struct drm_device *drm = dev_get_drvdata(dev);
1353 return drm_mode_config_helper_resume(drm);
1357 static SIMPLE_DEV_PM_OPS(host1x_drm_pm_ops, host1x_drm_suspend,
1360 static const struct of_device_id host1x_drm_subdevs[] = {
1361 { .compatible = "nvidia,tegra20-dc", },
1362 { .compatible = "nvidia,tegra20-hdmi", },
1363 { .compatible = "nvidia,tegra20-gr2d", },
1364 { .compatible = "nvidia,tegra20-gr3d", },
1365 { .compatible = "nvidia,tegra30-dc", },
1366 { .compatible = "nvidia,tegra30-hdmi", },
1367 { .compatible = "nvidia,tegra30-gr2d", },
1368 { .compatible = "nvidia,tegra30-gr3d", },
1369 { .compatible = "nvidia,tegra114-dc", },
1370 { .compatible = "nvidia,tegra114-dsi", },
1371 { .compatible = "nvidia,tegra114-hdmi", },
1372 { .compatible = "nvidia,tegra114-gr2d", },
1373 { .compatible = "nvidia,tegra114-gr3d", },
1374 { .compatible = "nvidia,tegra124-dc", },
1375 { .compatible = "nvidia,tegra124-sor", },
1376 { .compatible = "nvidia,tegra124-hdmi", },
1377 { .compatible = "nvidia,tegra124-dsi", },
1378 { .compatible = "nvidia,tegra124-vic", },
1379 { .compatible = "nvidia,tegra132-dsi", },
1380 { .compatible = "nvidia,tegra210-dc", },
1381 { .compatible = "nvidia,tegra210-dsi", },
1382 { .compatible = "nvidia,tegra210-sor", },
1383 { .compatible = "nvidia,tegra210-sor1", },
1384 { .compatible = "nvidia,tegra210-vic", },
1385 { .compatible = "nvidia,tegra210-nvdec", },
1386 { .compatible = "nvidia,tegra186-display", },
1387 { .compatible = "nvidia,tegra186-dc", },
1388 { .compatible = "nvidia,tegra186-sor", },
1389 { .compatible = "nvidia,tegra186-sor1", },
1390 { .compatible = "nvidia,tegra186-vic", },
1391 { .compatible = "nvidia,tegra186-nvdec", },
1392 { .compatible = "nvidia,tegra194-display", },
1393 { .compatible = "nvidia,tegra194-dc", },
1394 { .compatible = "nvidia,tegra194-sor", },
1395 { .compatible = "nvidia,tegra194-vic", },
1396 { .compatible = "nvidia,tegra194-nvdec", },
1397 { .compatible = "nvidia,tegra234-vic", },
1398 { .compatible = "nvidia,tegra234-nvdec", },
1402 static struct host1x_driver host1x_drm_driver = {
1405 .pm = &host1x_drm_pm_ops,
1407 .probe = host1x_drm_probe,
1408 .remove = host1x_drm_remove,
1409 .shutdown = host1x_drm_shutdown,
1410 .subdevs = host1x_drm_subdevs,
1413 static struct platform_driver * const drivers[] = {
1414 &tegra_display_hub_driver,
1418 &tegra_dpaux_driver,
1423 &tegra_nvdec_driver,
1426 static int __init host1x_drm_init(void)
1430 if (drm_firmware_drivers_only())
1433 err = host1x_driver_register(&host1x_drm_driver);
1437 err = platform_register_drivers(drivers, ARRAY_SIZE(drivers));
1439 goto unregister_host1x;
1444 host1x_driver_unregister(&host1x_drm_driver);
1447 module_init(host1x_drm_init);
1449 static void __exit host1x_drm_exit(void)
1451 platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
1452 host1x_driver_unregister(&host1x_drm_driver);
1454 module_exit(host1x_drm_exit);
1457 MODULE_DESCRIPTION("NVIDIA Tegra DRM driver");
1458 MODULE_LICENSE("GPL v2");