1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * NVIDIA Tegra Video decoder driver
4 *
5 * Copyright (C) 2016-2019 GRATE-DRIVER project
6 */
7
8 #include <linux/dma-buf.h>
9 #include <linux/iova.h>
10 #include <linux/kernel.h>
11 #include <linux/list.h>
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/workqueue.h>
15
16 #include "vde.h"
17
18 struct tegra_vde_cache_entry {
19 enum dma_data_direction dma_dir;
20 struct dma_buf_attachment *a;
21 struct delayed_work dwork;
22 struct tegra_vde *vde;
23 struct list_head list;
24 struct sg_table *sgt;
25 struct iova *iova;
26 unsigned int refcnt;
27 };
28
tegra_vde_release_entry(struct tegra_vde_cache_entry * entry)29 static void tegra_vde_release_entry(struct tegra_vde_cache_entry *entry)
30 {
31 struct dma_buf *dmabuf = entry->a->dmabuf;
32
33 WARN_ON_ONCE(entry->refcnt);
34
35 if (entry->vde->domain)
36 tegra_vde_iommu_unmap(entry->vde, entry->iova);
37
38 dma_buf_unmap_attachment(entry->a, entry->sgt, entry->dma_dir);
39 dma_buf_detach(dmabuf, entry->a);
40 dma_buf_put(dmabuf);
41
42 list_del(&entry->list);
43 kfree(entry);
44 }
45
tegra_vde_delayed_unmap(struct work_struct * work)46 static void tegra_vde_delayed_unmap(struct work_struct *work)
47 {
48 struct tegra_vde_cache_entry *entry;
49 struct tegra_vde *vde;
50
51 entry = container_of(work, struct tegra_vde_cache_entry,
52 dwork.work);
53 vde = entry->vde;
54
55 mutex_lock(&vde->map_lock);
56 tegra_vde_release_entry(entry);
57 mutex_unlock(&vde->map_lock);
58 }
59
tegra_vde_dmabuf_cache_map(struct tegra_vde * vde,struct dma_buf * dmabuf,enum dma_data_direction dma_dir,struct dma_buf_attachment ** ap,dma_addr_t * addrp)60 int tegra_vde_dmabuf_cache_map(struct tegra_vde *vde,
61 struct dma_buf *dmabuf,
62 enum dma_data_direction dma_dir,
63 struct dma_buf_attachment **ap,
64 dma_addr_t *addrp)
65 {
66 struct device *dev = vde->miscdev.parent;
67 struct dma_buf_attachment *attachment;
68 struct tegra_vde_cache_entry *entry;
69 struct sg_table *sgt;
70 struct iova *iova;
71 int err;
72
73 mutex_lock(&vde->map_lock);
74
75 list_for_each_entry(entry, &vde->map_list, list) {
76 if (entry->a->dmabuf != dmabuf)
77 continue;
78
79 if (!cancel_delayed_work(&entry->dwork))
80 continue;
81
82 if (entry->dma_dir != dma_dir)
83 entry->dma_dir = DMA_BIDIRECTIONAL;
84
85 dma_buf_put(dmabuf);
86
87 if (vde->domain)
88 *addrp = iova_dma_addr(&vde->iova, entry->iova);
89 else
90 *addrp = sg_dma_address(entry->sgt->sgl);
91
92 goto ref;
93 }
94
95 attachment = dma_buf_attach(dmabuf, dev);
96 if (IS_ERR(attachment)) {
97 dev_err(dev, "Failed to attach dmabuf\n");
98 err = PTR_ERR(attachment);
99 goto err_unlock;
100 }
101
102 sgt = dma_buf_map_attachment(attachment, dma_dir);
103 if (IS_ERR(sgt)) {
104 dev_err(dev, "Failed to get dmabufs sg_table\n");
105 err = PTR_ERR(sgt);
106 goto err_detach;
107 }
108
109 if (!vde->domain && sgt->nents > 1) {
110 dev_err(dev, "Sparse DMA region is unsupported, please enable IOMMU\n");
111 err = -EINVAL;
112 goto err_unmap;
113 }
114
115 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
116 if (!entry) {
117 err = -ENOMEM;
118 goto err_unmap;
119 }
120
121 if (vde->domain) {
122 err = tegra_vde_iommu_map(vde, sgt, &iova, dmabuf->size);
123 if (err)
124 goto err_free;
125
126 *addrp = iova_dma_addr(&vde->iova, iova);
127 } else {
128 *addrp = sg_dma_address(sgt->sgl);
129 iova = NULL;
130 }
131
132 INIT_DELAYED_WORK(&entry->dwork, tegra_vde_delayed_unmap);
133 list_add(&entry->list, &vde->map_list);
134
135 entry->dma_dir = dma_dir;
136 entry->iova = iova;
137 entry->vde = vde;
138 entry->sgt = sgt;
139 entry->a = attachment;
140 ref:
141 entry->refcnt++;
142
143 *ap = entry->a;
144
145 mutex_unlock(&vde->map_lock);
146
147 return 0;
148
149 err_free:
150 kfree(entry);
151 err_unmap:
152 dma_buf_unmap_attachment(attachment, sgt, dma_dir);
153 err_detach:
154 dma_buf_detach(dmabuf, attachment);
155 err_unlock:
156 mutex_unlock(&vde->map_lock);
157
158 return err;
159 }
160
tegra_vde_dmabuf_cache_unmap(struct tegra_vde * vde,struct dma_buf_attachment * a,bool release)161 void tegra_vde_dmabuf_cache_unmap(struct tegra_vde *vde,
162 struct dma_buf_attachment *a,
163 bool release)
164 {
165 struct tegra_vde_cache_entry *entry;
166
167 mutex_lock(&vde->map_lock);
168
169 list_for_each_entry(entry, &vde->map_list, list) {
170 if (entry->a != a)
171 continue;
172
173 WARN_ON_ONCE(!entry->refcnt);
174
175 if (--entry->refcnt == 0) {
176 if (release)
177 tegra_vde_release_entry(entry);
178 else
179 schedule_delayed_work(&entry->dwork, 5 * HZ);
180 }
181 break;
182 }
183
184 mutex_unlock(&vde->map_lock);
185 }
186
tegra_vde_dmabuf_cache_unmap_sync(struct tegra_vde * vde)187 void tegra_vde_dmabuf_cache_unmap_sync(struct tegra_vde *vde)
188 {
189 struct tegra_vde_cache_entry *entry, *tmp;
190
191 mutex_lock(&vde->map_lock);
192
193 list_for_each_entry_safe(entry, tmp, &vde->map_list, list) {
194 if (entry->refcnt)
195 continue;
196
197 if (!cancel_delayed_work(&entry->dwork))
198 continue;
199
200 tegra_vde_release_entry(entry);
201 }
202
203 mutex_unlock(&vde->map_lock);
204 }
205
tegra_vde_dmabuf_cache_unmap_all(struct tegra_vde * vde)206 void tegra_vde_dmabuf_cache_unmap_all(struct tegra_vde *vde)
207 {
208 struct tegra_vde_cache_entry *entry, *tmp;
209
210 mutex_lock(&vde->map_lock);
211
212 while (!list_empty(&vde->map_list)) {
213 list_for_each_entry_safe(entry, tmp, &vde->map_list, list) {
214 if (!cancel_delayed_work(&entry->dwork))
215 continue;
216
217 tegra_vde_release_entry(entry);
218 }
219
220 mutex_unlock(&vde->map_lock);
221 schedule();
222 mutex_lock(&vde->map_lock);
223 }
224
225 mutex_unlock(&vde->map_lock);
226 }
227