diff options
author | Alex Waterman <alexw@nvidia.com> | 2017-09-07 18:27:55 -0400 |
---|---|---|
committer | mobile promotions <svcmobile_promotions@nvidia.com> | 2017-10-18 19:00:37 -0400 |
commit | b3446bc0b6fca6cb992667f80a95f8503b6a652a (patch) | |
tree | 9882c36bfaef83da9d0a6eefec5e8c3564b93cea /drivers/gpu/nvgpu | |
parent | bee9c830c7898ceebf8c396b40598350229a7203 (diff) |
gpu: nvgpu: Move dma_buf usage from mm_gk20a.c
Move most of the dma_buf usage present in the mm_gk20a.c code
out to Linux specific code and some commom/mm code. There's
two primary groups of code:
1. dma_buf priv field code (for holding comptag data)
2. Comptag usage that relies on dma_buf pointers
For (1) the dma_buf code was simply moved to common/linux/dmabuf.c
since most of this code is clearly Linux specific.
The comptag code was a bit more complicated since there is two
parts to the comptag code. Firstly there's the code that manages
the comptag memory. This is essentially a simple allocator. This
was moved to common/mm/comptags.c since it can be shared across
all chips. The second set of code is moved to
common/linux/comptags.c since it is the interface between dma_bufs
and the comptag memory.
Two other fixes were done as well:
- Add struct gk20a to the comptag allocator init so that
the proper nvgpu_vzalloc() function could be used.
- Add necessary includes to common/linux/vm_priv.h.
JIRA NVGPU-30
JIRA NVGPU-138
Change-Id: I96c57f2763e5ebe18a2f2ee4b33e0e1a2597848c
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1566628
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu')
-rw-r--r-- | drivers/gpu/nvgpu/Makefile.nvgpu | 3 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/cde.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/comptags.c | 70 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/dmabuf.c | 247 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/dmabuf.h | 73 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/vm.c | 29 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/vm_priv.h | 5 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/mm/comptags.c | 95 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/gr_gk20a.c | 32 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/gr_gk20a.h | 14 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/mm_gk20a.c | 298 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/mm_gk20a.h | 19 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/gm20b/ltc_gm20b.c | 2 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/gp10b/ltc_gp10b.c | 2 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/include/nvgpu/comptags.h | 57 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/vgpu/gr_vgpu.c | 2 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/vgpu/ltc_vgpu.c | 2 |
17 files changed, 556 insertions, 395 deletions
diff --git a/drivers/gpu/nvgpu/Makefile.nvgpu b/drivers/gpu/nvgpu/Makefile.nvgpu index 42447e0c..81b0d1b8 100644 --- a/drivers/gpu/nvgpu/Makefile.nvgpu +++ b/drivers/gpu/nvgpu/Makefile.nvgpu | |||
@@ -51,6 +51,8 @@ nvgpu-y := \ | |||
51 | common/linux/rwsem.o \ | 51 | common/linux/rwsem.o \ |
52 | common/linux/cde_gm20b.o \ | 52 | common/linux/cde_gm20b.o \ |
53 | common/linux/cde_gp10b.o \ | 53 | common/linux/cde_gp10b.o \ |
54 | common/linux/comptags.o \ | ||
55 | common/linux/dmabuf.o \ | ||
54 | common/mm/nvgpu_allocator.o \ | 56 | common/mm/nvgpu_allocator.o \ |
55 | common/mm/bitmap_allocator.o \ | 57 | common/mm/bitmap_allocator.o \ |
56 | common/mm/buddy_allocator.o \ | 58 | common/mm/buddy_allocator.o \ |
@@ -61,6 +63,7 @@ nvgpu-y := \ | |||
61 | common/mm/vm.o \ | 63 | common/mm/vm.o \ |
62 | common/mm/vm_area.o \ | 64 | common/mm/vm_area.o \ |
63 | common/mm/nvgpu_mem.o \ | 65 | common/mm/nvgpu_mem.o \ |
66 | common/mm/comptags.o \ | ||
64 | common/bus.o \ | 67 | common/bus.o \ |
65 | common/enabled.o \ | 68 | common/enabled.o \ |
66 | common/pramin.o \ | 69 | common/pramin.o \ |
diff --git a/drivers/gpu/nvgpu/common/linux/cde.c b/drivers/gpu/nvgpu/common/linux/cde.c index c3a9b770..577d86e8 100644 --- a/drivers/gpu/nvgpu/common/linux/cde.c +++ b/drivers/gpu/nvgpu/common/linux/cde.c | |||
@@ -39,6 +39,7 @@ | |||
39 | 39 | ||
40 | #include "cde.h" | 40 | #include "cde.h" |
41 | #include "os_linux.h" | 41 | #include "os_linux.h" |
42 | #include "dmabuf.h" | ||
42 | 43 | ||
43 | #include <nvgpu/hw/gk20a/hw_ccsr_gk20a.h> | 44 | #include <nvgpu/hw/gk20a/hw_ccsr_gk20a.h> |
44 | #include <nvgpu/hw/gk20a/hw_pbdma_gk20a.h> | 45 | #include <nvgpu/hw/gk20a/hw_pbdma_gk20a.h> |
diff --git a/drivers/gpu/nvgpu/common/linux/comptags.c b/drivers/gpu/nvgpu/common/linux/comptags.c new file mode 100644 index 00000000..517429d8 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/comptags.c | |||
@@ -0,0 +1,70 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms and conditions of the GNU General Public License, | ||
6 | * version 2, as published by the Free Software Foundation. | ||
7 | * | ||
8 | * This program is distributed in the hope 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 | ||
11 | * more details. | ||
12 | * | ||
13 | * You should have received a copy of the GNU General Public License | ||
14 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
15 | */ | ||
16 | |||
17 | #include <linux/dma-buf.h> | ||
18 | |||
19 | #include <nvgpu/comptags.h> | ||
20 | |||
21 | #include "dmabuf.h" | ||
22 | |||
23 | void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf, | ||
24 | struct gk20a_comptags *comptags) | ||
25 | { | ||
26 | struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); | ||
27 | |||
28 | if (!comptags) | ||
29 | return; | ||
30 | |||
31 | if (!priv) { | ||
32 | memset(comptags, 0, sizeof(*comptags)); | ||
33 | return; | ||
34 | } | ||
35 | |||
36 | *comptags = priv->comptags; | ||
37 | } | ||
38 | |||
39 | int gk20a_alloc_comptags(struct gk20a *g, | ||
40 | struct device *dev, | ||
41 | struct dma_buf *dmabuf, | ||
42 | struct gk20a_comptag_allocator *allocator, | ||
43 | u32 lines) | ||
44 | { | ||
45 | struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); | ||
46 | u32 ctaglines_allocsize; | ||
47 | u32 offset; | ||
48 | int err; | ||
49 | |||
50 | if (!priv) | ||
51 | return -ENOSYS; | ||
52 | |||
53 | if (!lines) | ||
54 | return -EINVAL; | ||
55 | |||
56 | ctaglines_allocsize = lines; | ||
57 | |||
58 | /* store the allocator so we can use it when we free the ctags */ | ||
59 | priv->comptag_allocator = allocator; | ||
60 | err = gk20a_comptaglines_alloc(allocator, &offset, | ||
61 | ctaglines_allocsize); | ||
62 | if (err) | ||
63 | return err; | ||
64 | |||
65 | priv->comptags.offset = offset; | ||
66 | priv->comptags.lines = lines; | ||
67 | priv->comptags.allocated_lines = ctaglines_allocsize; | ||
68 | |||
69 | return 0; | ||
70 | } | ||
diff --git a/drivers/gpu/nvgpu/common/linux/dmabuf.c b/drivers/gpu/nvgpu/common/linux/dmabuf.c new file mode 100644 index 00000000..0b07b255 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/dmabuf.c | |||
@@ -0,0 +1,247 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms and conditions of the GNU General Public License, | ||
6 | * version 2, as published by the Free Software Foundation. | ||
7 | * | ||
8 | * This program is distributed in the hope 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 | ||
11 | * more details. | ||
12 | * | ||
13 | * You should have received a copy of the GNU General Public License | ||
14 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
15 | */ | ||
16 | |||
17 | #include <linux/device.h> | ||
18 | #include <linux/dma-buf.h> | ||
19 | #include <linux/scatterlist.h> | ||
20 | |||
21 | #include <nvgpu/comptags.h> | ||
22 | #include <nvgpu/enabled.h> | ||
23 | |||
24 | #include <nvgpu/linux/vidmem.h> | ||
25 | |||
26 | #include "gk20a/gk20a.h" | ||
27 | #include "gk20a/platform_gk20a.h" | ||
28 | |||
29 | #include "dmabuf.h" | ||
30 | #include "vm_priv.h" | ||
31 | #include "os_linux.h" | ||
32 | |||
33 | static void gk20a_mm_delete_priv(void *_priv) | ||
34 | { | ||
35 | struct gk20a_buffer_state *s, *s_tmp; | ||
36 | struct gk20a_dmabuf_priv *priv = _priv; | ||
37 | struct gk20a *g; | ||
38 | |||
39 | if (!priv) | ||
40 | return; | ||
41 | |||
42 | g = priv->g; | ||
43 | |||
44 | if (priv->comptags.lines) { | ||
45 | BUG_ON(!priv->comptag_allocator); | ||
46 | gk20a_comptaglines_free(priv->comptag_allocator, | ||
47 | priv->comptags.offset, | ||
48 | priv->comptags.allocated_lines); | ||
49 | } | ||
50 | |||
51 | /* Free buffer states */ | ||
52 | nvgpu_list_for_each_entry_safe(s, s_tmp, &priv->states, | ||
53 | gk20a_buffer_state, list) { | ||
54 | gk20a_fence_put(s->fence); | ||
55 | nvgpu_list_del(&s->list); | ||
56 | nvgpu_kfree(g, s); | ||
57 | } | ||
58 | |||
59 | nvgpu_kfree(g, priv); | ||
60 | } | ||
61 | |||
62 | enum nvgpu_aperture gk20a_dmabuf_aperture(struct gk20a *g, | ||
63 | struct dma_buf *dmabuf) | ||
64 | { | ||
65 | struct gk20a *buf_owner = nvgpu_vidmem_buf_owner(dmabuf); | ||
66 | bool unified_memory = nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY); | ||
67 | |||
68 | if (buf_owner == NULL) { | ||
69 | /* Not nvgpu-allocated, assume system memory */ | ||
70 | return APERTURE_SYSMEM; | ||
71 | } else if (WARN_ON(buf_owner == g && unified_memory)) { | ||
72 | /* Looks like our video memory, but this gpu doesn't support | ||
73 | * it. Warn about a bug and bail out */ | ||
74 | nvgpu_warn(g, | ||
75 | "dmabuf is our vidmem but we don't have local vidmem"); | ||
76 | return APERTURE_INVALID; | ||
77 | } else if (buf_owner != g) { | ||
78 | /* Someone else's vidmem */ | ||
79 | return APERTURE_INVALID; | ||
80 | } else { | ||
81 | /* Yay, buf_owner == g */ | ||
82 | return APERTURE_VIDMEM; | ||
83 | } | ||
84 | } | ||
85 | |||
86 | struct sg_table *gk20a_mm_pin(struct device *dev, struct dma_buf *dmabuf) | ||
87 | { | ||
88 | struct gk20a_dmabuf_priv *priv; | ||
89 | |||
90 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
91 | if (WARN_ON(!priv)) | ||
92 | return ERR_PTR(-EINVAL); | ||
93 | |||
94 | nvgpu_mutex_acquire(&priv->lock); | ||
95 | |||
96 | if (priv->pin_count == 0) { | ||
97 | priv->attach = dma_buf_attach(dmabuf, dev); | ||
98 | if (IS_ERR(priv->attach)) { | ||
99 | nvgpu_mutex_release(&priv->lock); | ||
100 | return (struct sg_table *)priv->attach; | ||
101 | } | ||
102 | |||
103 | priv->sgt = dma_buf_map_attachment(priv->attach, | ||
104 | DMA_BIDIRECTIONAL); | ||
105 | if (IS_ERR(priv->sgt)) { | ||
106 | dma_buf_detach(dmabuf, priv->attach); | ||
107 | nvgpu_mutex_release(&priv->lock); | ||
108 | return priv->sgt; | ||
109 | } | ||
110 | } | ||
111 | |||
112 | priv->pin_count++; | ||
113 | nvgpu_mutex_release(&priv->lock); | ||
114 | return priv->sgt; | ||
115 | } | ||
116 | |||
117 | void gk20a_mm_unpin(struct device *dev, struct dma_buf *dmabuf, | ||
118 | struct sg_table *sgt) | ||
119 | { | ||
120 | struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); | ||
121 | dma_addr_t dma_addr; | ||
122 | |||
123 | if (IS_ERR(priv) || !priv) | ||
124 | return; | ||
125 | |||
126 | nvgpu_mutex_acquire(&priv->lock); | ||
127 | WARN_ON(priv->sgt != sgt); | ||
128 | priv->pin_count--; | ||
129 | WARN_ON(priv->pin_count < 0); | ||
130 | dma_addr = sg_dma_address(priv->sgt->sgl); | ||
131 | if (priv->pin_count == 0) { | ||
132 | dma_buf_unmap_attachment(priv->attach, priv->sgt, | ||
133 | DMA_BIDIRECTIONAL); | ||
134 | dma_buf_detach(dmabuf, priv->attach); | ||
135 | } | ||
136 | nvgpu_mutex_release(&priv->lock); | ||
137 | } | ||
138 | |||
139 | int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev) | ||
140 | { | ||
141 | struct gk20a *g = gk20a_get_platform(dev)->g; | ||
142 | struct gk20a_dmabuf_priv *priv; | ||
143 | static u64 priv_count = 0; | ||
144 | |||
145 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
146 | if (likely(priv)) | ||
147 | return 0; | ||
148 | |||
149 | nvgpu_mutex_acquire(&g->mm.priv_lock); | ||
150 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
151 | if (priv) | ||
152 | goto priv_exist_or_err; | ||
153 | |||
154 | priv = nvgpu_kzalloc(g, sizeof(*priv)); | ||
155 | if (!priv) { | ||
156 | priv = ERR_PTR(-ENOMEM); | ||
157 | goto priv_exist_or_err; | ||
158 | } | ||
159 | |||
160 | nvgpu_mutex_init(&priv->lock); | ||
161 | nvgpu_init_list_node(&priv->states); | ||
162 | priv->buffer_id = ++priv_count; | ||
163 | priv->g = g; | ||
164 | dma_buf_set_drvdata(dmabuf, dev, priv, gk20a_mm_delete_priv); | ||
165 | |||
166 | priv_exist_or_err: | ||
167 | nvgpu_mutex_release(&g->mm.priv_lock); | ||
168 | if (IS_ERR(priv)) | ||
169 | return -ENOMEM; | ||
170 | |||
171 | return 0; | ||
172 | } | ||
173 | |||
174 | int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g, | ||
175 | u64 offset, struct gk20a_buffer_state **state) | ||
176 | { | ||
177 | int err = 0; | ||
178 | struct gk20a_dmabuf_priv *priv; | ||
179 | struct gk20a_buffer_state *s; | ||
180 | struct device *dev = dev_from_gk20a(g); | ||
181 | |||
182 | if (WARN_ON(offset >= (u64)dmabuf->size)) | ||
183 | return -EINVAL; | ||
184 | |||
185 | err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev); | ||
186 | if (err) | ||
187 | return err; | ||
188 | |||
189 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
190 | if (WARN_ON(!priv)) | ||
191 | return -ENOSYS; | ||
192 | |||
193 | nvgpu_mutex_acquire(&priv->lock); | ||
194 | |||
195 | nvgpu_list_for_each_entry(s, &priv->states, gk20a_buffer_state, list) | ||
196 | if (s->offset == offset) | ||
197 | goto out; | ||
198 | |||
199 | /* State not found, create state. */ | ||
200 | s = nvgpu_kzalloc(g, sizeof(*s)); | ||
201 | if (!s) { | ||
202 | err = -ENOMEM; | ||
203 | goto out; | ||
204 | } | ||
205 | |||
206 | s->offset = offset; | ||
207 | nvgpu_init_list_node(&s->list); | ||
208 | nvgpu_mutex_init(&s->lock); | ||
209 | nvgpu_list_add_tail(&s->list, &priv->states); | ||
210 | |||
211 | out: | ||
212 | nvgpu_mutex_release(&priv->lock); | ||
213 | if (!err) | ||
214 | *state = s; | ||
215 | return err; | ||
216 | } | ||
217 | |||
218 | int gk20a_mm_get_buffer_info(struct device *dev, int dmabuf_fd, | ||
219 | u64 *buffer_id, u64 *buffer_len) | ||
220 | { | ||
221 | struct dma_buf *dmabuf; | ||
222 | struct gk20a_dmabuf_priv *priv; | ||
223 | int err = 0; | ||
224 | |||
225 | dmabuf = dma_buf_get(dmabuf_fd); | ||
226 | if (IS_ERR(dmabuf)) { | ||
227 | dev_warn(dev, "%s: fd %d is not a dmabuf", __func__, dmabuf_fd); | ||
228 | return PTR_ERR(dmabuf); | ||
229 | } | ||
230 | |||
231 | err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev); | ||
232 | if (err) { | ||
233 | dev_warn(dev, "Failed to allocate dmabuf drvdata (err = %d)", | ||
234 | err); | ||
235 | goto clean_up; | ||
236 | } | ||
237 | |||
238 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
239 | if (likely(priv)) { | ||
240 | *buffer_id = priv->buffer_id; | ||
241 | *buffer_len = dmabuf->size; | ||
242 | } | ||
243 | |||
244 | clean_up: | ||
245 | dma_buf_put(dmabuf); | ||
246 | return err; | ||
247 | } | ||
diff --git a/drivers/gpu/nvgpu/common/linux/dmabuf.h b/drivers/gpu/nvgpu/common/linux/dmabuf.h new file mode 100644 index 00000000..718386c5 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/dmabuf.h | |||
@@ -0,0 +1,73 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms and conditions of the GNU General Public License, | ||
6 | * version 2, as published by the Free Software Foundation. | ||
7 | * | ||
8 | * This program is distributed in the hope 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 | ||
11 | * more details. | ||
12 | * | ||
13 | * You should have received a copy of the GNU General Public License | ||
14 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
15 | */ | ||
16 | |||
17 | #ifndef __COMMON_LINUX_DMABUF_H__ | ||
18 | #define __COMMON_LINUX_DMABUF_H__ | ||
19 | |||
20 | #include <nvgpu/comptags.h> | ||
21 | #include <nvgpu/list.h> | ||
22 | #include <nvgpu/lock.h> | ||
23 | #include <nvgpu/gmmu.h> | ||
24 | |||
25 | struct sg_table; | ||
26 | struct dma_buf; | ||
27 | struct dma_buf_attachment; | ||
28 | struct device; | ||
29 | |||
30 | struct gk20a; | ||
31 | struct gk20a_buffer_state; | ||
32 | |||
33 | struct gk20a_dmabuf_priv { | ||
34 | struct nvgpu_mutex lock; | ||
35 | |||
36 | struct gk20a *g; | ||
37 | |||
38 | struct gk20a_comptag_allocator *comptag_allocator; | ||
39 | struct gk20a_comptags comptags; | ||
40 | |||
41 | struct dma_buf_attachment *attach; | ||
42 | struct sg_table *sgt; | ||
43 | |||
44 | int pin_count; | ||
45 | |||
46 | struct nvgpu_list_node states; | ||
47 | |||
48 | u64 buffer_id; | ||
49 | }; | ||
50 | |||
51 | /* | ||
52 | * These are implemented in common/linux/comptags.c - these are dmabuf related | ||
53 | * functions though so they are defined here. They cannot be defined in | ||
54 | * <nvgpu/comptags.h> since that file must be OS agnostic. | ||
55 | */ | ||
56 | int gk20a_alloc_comptags(struct gk20a *g, | ||
57 | struct device *dev, | ||
58 | struct dma_buf *dmabuf, | ||
59 | struct gk20a_comptag_allocator *allocator, | ||
60 | u32 lines); | ||
61 | void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf, | ||
62 | struct gk20a_comptags *comptags); | ||
63 | |||
64 | struct sg_table *gk20a_mm_pin(struct device *dev, struct dma_buf *dmabuf); | ||
65 | void gk20a_mm_unpin(struct device *dev, struct dma_buf *dmabuf, | ||
66 | struct sg_table *sgt); | ||
67 | |||
68 | int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev); | ||
69 | |||
70 | int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g, | ||
71 | u64 offset, struct gk20a_buffer_state **state); | ||
72 | |||
73 | #endif | ||
diff --git a/drivers/gpu/nvgpu/common/linux/vm.c b/drivers/gpu/nvgpu/common/linux/vm.c index 45058321..b686d616 100644 --- a/drivers/gpu/nvgpu/common/linux/vm.c +++ b/drivers/gpu/nvgpu/common/linux/vm.c | |||
@@ -24,7 +24,6 @@ | |||
24 | #include <nvgpu/nvgpu_mem.h> | 24 | #include <nvgpu/nvgpu_mem.h> |
25 | #include <nvgpu/page_allocator.h> | 25 | #include <nvgpu/page_allocator.h> |
26 | #include <nvgpu/vidmem.h> | 26 | #include <nvgpu/vidmem.h> |
27 | #include <nvgpu/enabled.h> | ||
28 | 27 | ||
29 | #include <nvgpu/linux/vidmem.h> | 28 | #include <nvgpu/linux/vidmem.h> |
30 | #include <nvgpu/linux/nvgpu_mem.h> | 29 | #include <nvgpu/linux/nvgpu_mem.h> |
@@ -36,33 +35,7 @@ | |||
36 | 35 | ||
37 | #include "vm_priv.h" | 36 | #include "vm_priv.h" |
38 | #include "os_linux.h" | 37 | #include "os_linux.h" |
39 | 38 | #include "dmabuf.h" | |
40 | /* | ||
41 | * Temporary location for this code until a dmabuf.c file exists. | ||
42 | */ | ||
43 | enum nvgpu_aperture gk20a_dmabuf_aperture(struct gk20a *g, | ||
44 | struct dma_buf *dmabuf) | ||
45 | { | ||
46 | struct gk20a *buf_owner = nvgpu_vidmem_buf_owner(dmabuf); | ||
47 | bool unified_memory = nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY); | ||
48 | |||
49 | if (buf_owner == NULL) { | ||
50 | /* Not nvgpu-allocated, assume system memory */ | ||
51 | return APERTURE_SYSMEM; | ||
52 | } else if (WARN_ON(buf_owner == g && unified_memory)) { | ||
53 | /* Looks like our video memory, but this gpu doesn't support | ||
54 | * it. Warn about a bug and bail out */ | ||
55 | nvgpu_warn(g, | ||
56 | "dmabuf is our vidmem but we don't have local vidmem"); | ||
57 | return APERTURE_INVALID; | ||
58 | } else if (buf_owner != g) { | ||
59 | /* Someone else's vidmem */ | ||
60 | return APERTURE_INVALID; | ||
61 | } else { | ||
62 | /* Yay, buf_owner == g */ | ||
63 | return APERTURE_VIDMEM; | ||
64 | } | ||
65 | } | ||
66 | 39 | ||
67 | static struct nvgpu_mapped_buf *__nvgpu_vm_find_mapped_buf_reverse( | 40 | static struct nvgpu_mapped_buf *__nvgpu_vm_find_mapped_buf_reverse( |
68 | struct vm_gk20a *vm, struct dma_buf *dmabuf, u32 kind) | 41 | struct vm_gk20a *vm, struct dma_buf *dmabuf, u32 kind) |
diff --git a/drivers/gpu/nvgpu/common/linux/vm_priv.h b/drivers/gpu/nvgpu/common/linux/vm_priv.h index 1eadf1d0..fa173d59 100644 --- a/drivers/gpu/nvgpu/common/linux/vm_priv.h +++ b/drivers/gpu/nvgpu/common/linux/vm_priv.h | |||
@@ -102,10 +102,5 @@ int setup_buffer_kind_and_compression(struct vm_gk20a *vm, | |||
102 | u32 flags, | 102 | u32 flags, |
103 | struct buffer_attrs *bfr, | 103 | struct buffer_attrs *bfr, |
104 | enum gmmu_pgsz_gk20a pgsz_idx); | 104 | enum gmmu_pgsz_gk20a pgsz_idx); |
105 | int gk20a_alloc_comptags(struct gk20a *g, | ||
106 | struct device *dev, | ||
107 | struct dma_buf *dmabuf, | ||
108 | struct gk20a_comptag_allocator *allocator, | ||
109 | u32 lines); | ||
110 | 105 | ||
111 | #endif | 106 | #endif |
diff --git a/drivers/gpu/nvgpu/common/mm/comptags.c b/drivers/gpu/nvgpu/common/mm/comptags.c new file mode 100644 index 00000000..01ab646a --- /dev/null +++ b/drivers/gpu/nvgpu/common/mm/comptags.c | |||
@@ -0,0 +1,95 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. | ||
3 | * | ||
4 | * Permission is hereby granted, free of charge, to any person obtaining a | ||
5 | * copy of this software and associated documentation files (the "Software"), | ||
6 | * to deal in the Software without restriction, including without limitation | ||
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | ||
8 | * and/or sell copies of the Software, and to permit persons to whom the | ||
9 | * Software is furnished to do so, subject to the following conditions: | ||
10 | * | ||
11 | * The above copyright notice and this permission notice shall be included in | ||
12 | * all copies or substantial portions of the Software. | ||
13 | * | ||
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
17 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
18 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | ||
19 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER | ||
20 | * DEALINGS IN THE SOFTWARE. | ||
21 | */ | ||
22 | |||
23 | #include <nvgpu/bitops.h> | ||
24 | #include <nvgpu/comptags.h> | ||
25 | |||
26 | #include "gk20a/gk20a.h" | ||
27 | |||
28 | int gk20a_comptaglines_alloc(struct gk20a_comptag_allocator *allocator, | ||
29 | u32 *offset, u32 len) | ||
30 | { | ||
31 | unsigned long addr; | ||
32 | int err = 0; | ||
33 | |||
34 | nvgpu_mutex_acquire(&allocator->lock); | ||
35 | addr = bitmap_find_next_zero_area(allocator->bitmap, allocator->size, | ||
36 | 0, len, 0); | ||
37 | if (addr < allocator->size) { | ||
38 | /* number zero is reserved; bitmap base is 1 */ | ||
39 | *offset = 1 + addr; | ||
40 | bitmap_set(allocator->bitmap, addr, len); | ||
41 | } else { | ||
42 | err = -ENOMEM; | ||
43 | } | ||
44 | nvgpu_mutex_release(&allocator->lock); | ||
45 | |||
46 | return err; | ||
47 | } | ||
48 | |||
49 | void gk20a_comptaglines_free(struct gk20a_comptag_allocator *allocator, | ||
50 | u32 offset, u32 len) | ||
51 | { | ||
52 | /* number zero is reserved; bitmap base is 1 */ | ||
53 | u32 addr = offset - 1; | ||
54 | |||
55 | WARN_ON(offset == 0); | ||
56 | WARN_ON(addr > allocator->size); | ||
57 | WARN_ON(addr + len > allocator->size); | ||
58 | |||
59 | nvgpu_mutex_acquire(&allocator->lock); | ||
60 | bitmap_clear(allocator->bitmap, addr, len); | ||
61 | nvgpu_mutex_release(&allocator->lock); | ||
62 | } | ||
63 | |||
64 | int gk20a_comptag_allocator_init(struct gk20a *g, | ||
65 | struct gk20a_comptag_allocator *allocator, | ||
66 | unsigned long size) | ||
67 | { | ||
68 | nvgpu_mutex_init(&allocator->lock); | ||
69 | |||
70 | /* | ||
71 | * 0th comptag is special and is never used. The base for this bitmap | ||
72 | * is 1, and its size is one less than the size of comptag store. | ||
73 | */ | ||
74 | size--; | ||
75 | allocator->bitmap = nvgpu_vzalloc(g, | ||
76 | BITS_TO_LONGS(size) * sizeof(long)); | ||
77 | if (!allocator->bitmap) | ||
78 | return -ENOMEM; | ||
79 | |||
80 | allocator->size = size; | ||
81 | |||
82 | return 0; | ||
83 | } | ||
84 | |||
85 | void gk20a_comptag_allocator_destroy(struct gk20a *g, | ||
86 | struct gk20a_comptag_allocator *allocator) | ||
87 | { | ||
88 | /* | ||
89 | * called only when exiting the driver (gk20a_remove, or unwinding the | ||
90 | * init stage); no users should be active, so taking the mutex is | ||
91 | * unnecessary here. | ||
92 | */ | ||
93 | allocator->size = 0; | ||
94 | nvgpu_vfree(allocator->g, allocator->bitmap); | ||
95 | } | ||
diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c index 57c1c0bc..700dcdf8 100644 --- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c | |||
@@ -3044,36 +3044,6 @@ out: | |||
3044 | return err; | 3044 | return err; |
3045 | } | 3045 | } |
3046 | 3046 | ||
3047 | int gk20a_comptag_allocator_init(struct gk20a_comptag_allocator *allocator, | ||
3048 | unsigned long size) | ||
3049 | { | ||
3050 | nvgpu_mutex_init(&allocator->lock); | ||
3051 | /* | ||
3052 | * 0th comptag is special and is never used. The base for this bitmap | ||
3053 | * is 1, and its size is one less than the size of comptag store. | ||
3054 | */ | ||
3055 | size--; | ||
3056 | allocator->bitmap = vzalloc(BITS_TO_LONGS(size) * sizeof(long)); | ||
3057 | if (!allocator->bitmap) | ||
3058 | return -ENOMEM; | ||
3059 | allocator->size = size; | ||
3060 | return 0; | ||
3061 | } | ||
3062 | |||
3063 | void gk20a_comptag_allocator_destroy(struct gk20a_comptag_allocator *allocator) | ||
3064 | { | ||
3065 | struct gr_gk20a *gr = container_of(allocator, | ||
3066 | struct gr_gk20a, comp_tags); | ||
3067 | |||
3068 | /* | ||
3069 | * called only when exiting the driver (gk20a_remove, or unwinding the | ||
3070 | * init stage); no users should be active, so taking the mutex is | ||
3071 | * unnecessary here. | ||
3072 | */ | ||
3073 | allocator->size = 0; | ||
3074 | nvgpu_vfree(gr->g, allocator->bitmap); | ||
3075 | } | ||
3076 | |||
3077 | static void gk20a_remove_gr_support(struct gr_gk20a *gr) | 3047 | static void gk20a_remove_gr_support(struct gr_gk20a *gr) |
3078 | { | 3048 | { |
3079 | struct gk20a *g = gr->g; | 3049 | struct gk20a *g = gr->g; |
@@ -3148,7 +3118,7 @@ static void gk20a_remove_gr_support(struct gr_gk20a *gr) | |||
3148 | nvgpu_big_free(g, gr->ctx_vars.hwpm_ctxsw_buffer_offset_map); | 3118 | nvgpu_big_free(g, gr->ctx_vars.hwpm_ctxsw_buffer_offset_map); |
3149 | gr->ctx_vars.hwpm_ctxsw_buffer_offset_map = NULL; | 3119 | gr->ctx_vars.hwpm_ctxsw_buffer_offset_map = NULL; |
3150 | 3120 | ||
3151 | gk20a_comptag_allocator_destroy(&gr->comp_tags); | 3121 | gk20a_comptag_allocator_destroy(g, &gr->comp_tags); |
3152 | } | 3122 | } |
3153 | 3123 | ||
3154 | static int gr_gk20a_init_gr_config(struct gk20a *g, struct gr_gk20a *gr) | 3124 | static int gr_gk20a_init_gr_config(struct gk20a *g, struct gr_gk20a *gr) |
diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.h b/drivers/gpu/nvgpu/gk20a/gr_gk20a.h index 5fab43ca..84eb8970 100644 --- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.h | |||
@@ -36,6 +36,8 @@ | |||
36 | #include "gr_ctx_gk20a.h" | 36 | #include "gr_ctx_gk20a.h" |
37 | #include "mm_gk20a.h" | 37 | #include "mm_gk20a.h" |
38 | 38 | ||
39 | #include <nvgpu/comptags.h> | ||
40 | |||
39 | #define GR_IDLE_CHECK_DEFAULT 10 /* usec */ | 41 | #define GR_IDLE_CHECK_DEFAULT 10 /* usec */ |
40 | #define GR_IDLE_CHECK_MAX 200 /* usec */ | 42 | #define GR_IDLE_CHECK_MAX 200 /* usec */ |
41 | #define GR_FECS_POLL_INTERVAL 5 /* usec */ | 43 | #define GR_FECS_POLL_INTERVAL 5 /* usec */ |
@@ -342,13 +344,7 @@ struct gr_gk20a { | |||
342 | 344 | ||
343 | u32 max_comptag_mem; /* max memory size (MB) for comptag */ | 345 | u32 max_comptag_mem; /* max memory size (MB) for comptag */ |
344 | struct compbit_store_desc compbit_store; | 346 | struct compbit_store_desc compbit_store; |
345 | struct gk20a_comptag_allocator { | 347 | struct gk20a_comptag_allocator comp_tags; |
346 | struct nvgpu_mutex lock; | ||
347 | /* this bitmap starts at ctag 1. 0th cannot be taken */ | ||
348 | unsigned long *bitmap; | ||
349 | /* size of bitmap, not max ctags, so one less */ | ||
350 | unsigned long size; | ||
351 | } comp_tags; | ||
352 | 348 | ||
353 | struct gr_zcull_gk20a zcull; | 349 | struct gr_zcull_gk20a zcull; |
354 | 350 | ||
@@ -503,10 +499,6 @@ int gk20a_init_gr_support(struct gk20a *g); | |||
503 | int gk20a_enable_gr_hw(struct gk20a *g); | 499 | int gk20a_enable_gr_hw(struct gk20a *g); |
504 | int gk20a_gr_reset(struct gk20a *g); | 500 | int gk20a_gr_reset(struct gk20a *g); |
505 | void gk20a_gr_wait_initialized(struct gk20a *g); | 501 | void gk20a_gr_wait_initialized(struct gk20a *g); |
506 | /* real size here, but first (ctag 0) isn't used */ | ||
507 | int gk20a_comptag_allocator_init(struct gk20a_comptag_allocator *allocator, | ||
508 | unsigned long size); | ||
509 | void gk20a_comptag_allocator_destroy(struct gk20a_comptag_allocator *allocator); | ||
510 | 502 | ||
511 | int gk20a_init_gr_channel(struct channel_gk20a *ch_gk20a); | 503 | int gk20a_init_gr_channel(struct channel_gk20a *ch_gk20a); |
512 | 504 | ||
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c index 8936cd03..69d9e983 100644 --- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c | |||
@@ -22,7 +22,6 @@ | |||
22 | * DEALINGS IN THE SOFTWARE. | 22 | * DEALINGS IN THE SOFTWARE. |
23 | */ | 23 | */ |
24 | 24 | ||
25 | #include <linux/scatterlist.h> | ||
26 | #include <linux/dma-buf.h> | 25 | #include <linux/dma-buf.h> |
27 | #include <linux/dma-mapping.h> | 26 | #include <linux/dma-mapping.h> |
28 | #include <linux/dma-attrs.h> | 27 | #include <linux/dma-attrs.h> |
@@ -70,6 +69,7 @@ | |||
70 | * all the common APIs no longers have Linux stuff in them. | 69 | * all the common APIs no longers have Linux stuff in them. |
71 | */ | 70 | */ |
72 | #include "common/linux/vm_priv.h" | 71 | #include "common/linux/vm_priv.h" |
72 | #include "common/linux/dmabuf.h" | ||
73 | 73 | ||
74 | /* | 74 | /* |
75 | * GPU mapping life cycle | 75 | * GPU mapping life cycle |
@@ -108,190 +108,6 @@ static int __must_check gk20a_init_hwpm(struct mm_gk20a *mm); | |||
108 | static int __must_check gk20a_init_cde_vm(struct mm_gk20a *mm); | 108 | static int __must_check gk20a_init_cde_vm(struct mm_gk20a *mm); |
109 | static int __must_check gk20a_init_ce_vm(struct mm_gk20a *mm); | 109 | static int __must_check gk20a_init_ce_vm(struct mm_gk20a *mm); |
110 | 110 | ||
111 | struct gk20a_dmabuf_priv { | ||
112 | struct nvgpu_mutex lock; | ||
113 | |||
114 | struct gk20a *g; | ||
115 | |||
116 | struct gk20a_comptag_allocator *comptag_allocator; | ||
117 | struct gk20a_comptags comptags; | ||
118 | |||
119 | struct dma_buf_attachment *attach; | ||
120 | struct sg_table *sgt; | ||
121 | |||
122 | int pin_count; | ||
123 | |||
124 | struct nvgpu_list_node states; | ||
125 | |||
126 | u64 buffer_id; | ||
127 | }; | ||
128 | |||
129 | static int gk20a_comptaglines_alloc(struct gk20a_comptag_allocator *allocator, | ||
130 | u32 *offset, u32 len) | ||
131 | { | ||
132 | unsigned long addr; | ||
133 | int err = 0; | ||
134 | |||
135 | nvgpu_mutex_acquire(&allocator->lock); | ||
136 | addr = bitmap_find_next_zero_area(allocator->bitmap, allocator->size, | ||
137 | 0, len, 0); | ||
138 | if (addr < allocator->size) { | ||
139 | /* number zero is reserved; bitmap base is 1 */ | ||
140 | *offset = 1 + addr; | ||
141 | bitmap_set(allocator->bitmap, addr, len); | ||
142 | } else { | ||
143 | err = -ENOMEM; | ||
144 | } | ||
145 | nvgpu_mutex_release(&allocator->lock); | ||
146 | |||
147 | return err; | ||
148 | } | ||
149 | |||
150 | static void gk20a_comptaglines_free(struct gk20a_comptag_allocator *allocator, | ||
151 | u32 offset, u32 len) | ||
152 | { | ||
153 | /* number zero is reserved; bitmap base is 1 */ | ||
154 | u32 addr = offset - 1; | ||
155 | WARN_ON(offset == 0); | ||
156 | WARN_ON(addr > allocator->size); | ||
157 | WARN_ON(addr + len > allocator->size); | ||
158 | |||
159 | nvgpu_mutex_acquire(&allocator->lock); | ||
160 | bitmap_clear(allocator->bitmap, addr, len); | ||
161 | nvgpu_mutex_release(&allocator->lock); | ||
162 | } | ||
163 | |||
164 | static void gk20a_mm_delete_priv(void *_priv) | ||
165 | { | ||
166 | struct gk20a_buffer_state *s, *s_tmp; | ||
167 | struct gk20a_dmabuf_priv *priv = _priv; | ||
168 | struct gk20a *g; | ||
169 | |||
170 | if (!priv) | ||
171 | return; | ||
172 | |||
173 | g = priv->g; | ||
174 | |||
175 | if (priv->comptags.lines) { | ||
176 | BUG_ON(!priv->comptag_allocator); | ||
177 | gk20a_comptaglines_free(priv->comptag_allocator, | ||
178 | priv->comptags.offset, | ||
179 | priv->comptags.allocated_lines); | ||
180 | } | ||
181 | |||
182 | /* Free buffer states */ | ||
183 | nvgpu_list_for_each_entry_safe(s, s_tmp, &priv->states, | ||
184 | gk20a_buffer_state, list) { | ||
185 | gk20a_fence_put(s->fence); | ||
186 | nvgpu_list_del(&s->list); | ||
187 | nvgpu_kfree(g, s); | ||
188 | } | ||
189 | |||
190 | nvgpu_kfree(g, priv); | ||
191 | } | ||
192 | |||
193 | struct sg_table *gk20a_mm_pin(struct device *dev, struct dma_buf *dmabuf) | ||
194 | { | ||
195 | struct gk20a_dmabuf_priv *priv; | ||
196 | |||
197 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
198 | if (WARN_ON(!priv)) | ||
199 | return ERR_PTR(-EINVAL); | ||
200 | |||
201 | nvgpu_mutex_acquire(&priv->lock); | ||
202 | |||
203 | if (priv->pin_count == 0) { | ||
204 | priv->attach = dma_buf_attach(dmabuf, dev); | ||
205 | if (IS_ERR(priv->attach)) { | ||
206 | nvgpu_mutex_release(&priv->lock); | ||
207 | return (struct sg_table *)priv->attach; | ||
208 | } | ||
209 | |||
210 | priv->sgt = dma_buf_map_attachment(priv->attach, | ||
211 | DMA_BIDIRECTIONAL); | ||
212 | if (IS_ERR(priv->sgt)) { | ||
213 | dma_buf_detach(dmabuf, priv->attach); | ||
214 | nvgpu_mutex_release(&priv->lock); | ||
215 | return priv->sgt; | ||
216 | } | ||
217 | } | ||
218 | |||
219 | priv->pin_count++; | ||
220 | nvgpu_mutex_release(&priv->lock); | ||
221 | return priv->sgt; | ||
222 | } | ||
223 | |||
224 | void gk20a_mm_unpin(struct device *dev, struct dma_buf *dmabuf, | ||
225 | struct sg_table *sgt) | ||
226 | { | ||
227 | struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); | ||
228 | dma_addr_t dma_addr; | ||
229 | |||
230 | if (IS_ERR(priv) || !priv) | ||
231 | return; | ||
232 | |||
233 | nvgpu_mutex_acquire(&priv->lock); | ||
234 | WARN_ON(priv->sgt != sgt); | ||
235 | priv->pin_count--; | ||
236 | WARN_ON(priv->pin_count < 0); | ||
237 | dma_addr = sg_dma_address(priv->sgt->sgl); | ||
238 | if (priv->pin_count == 0) { | ||
239 | dma_buf_unmap_attachment(priv->attach, priv->sgt, | ||
240 | DMA_BIDIRECTIONAL); | ||
241 | dma_buf_detach(dmabuf, priv->attach); | ||
242 | } | ||
243 | nvgpu_mutex_release(&priv->lock); | ||
244 | } | ||
245 | |||
246 | void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf, | ||
247 | struct gk20a_comptags *comptags) | ||
248 | { | ||
249 | struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); | ||
250 | |||
251 | if (!comptags) | ||
252 | return; | ||
253 | |||
254 | if (!priv) { | ||
255 | memset(comptags, 0, sizeof(*comptags)); | ||
256 | return; | ||
257 | } | ||
258 | |||
259 | *comptags = priv->comptags; | ||
260 | } | ||
261 | |||
262 | int gk20a_alloc_comptags(struct gk20a *g, | ||
263 | struct device *dev, | ||
264 | struct dma_buf *dmabuf, | ||
265 | struct gk20a_comptag_allocator *allocator, | ||
266 | u32 lines) | ||
267 | { | ||
268 | struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); | ||
269 | u32 ctaglines_allocsize; | ||
270 | u32 offset; | ||
271 | int err; | ||
272 | |||
273 | if (!priv) | ||
274 | return -ENOSYS; | ||
275 | |||
276 | if (!lines) | ||
277 | return -EINVAL; | ||
278 | |||
279 | ctaglines_allocsize = lines; | ||
280 | |||
281 | /* store the allocator so we can use it when we free the ctags */ | ||
282 | priv->comptag_allocator = allocator; | ||
283 | err = gk20a_comptaglines_alloc(allocator, &offset, | ||
284 | ctaglines_allocsize); | ||
285 | if (err) | ||
286 | return err; | ||
287 | |||
288 | priv->comptags.offset = offset; | ||
289 | priv->comptags.lines = lines; | ||
290 | priv->comptags.allocated_lines = ctaglines_allocsize; | ||
291 | |||
292 | return 0; | ||
293 | } | ||
294 | |||
295 | static int gk20a_init_mm_reset_enable_hw(struct gk20a *g) | 111 | static int gk20a_init_mm_reset_enable_hw(struct gk20a *g) |
296 | { | 112 | { |
297 | gk20a_dbg_fn(""); | 113 | gk20a_dbg_fn(""); |
@@ -1037,87 +853,6 @@ int gk20a_vm_bind_channel(struct gk20a_as_share *as_share, | |||
1037 | return __gk20a_vm_bind_channel(as_share->vm, ch); | 853 | return __gk20a_vm_bind_channel(as_share->vm, ch); |
1038 | } | 854 | } |
1039 | 855 | ||
1040 | int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev) | ||
1041 | { | ||
1042 | struct gk20a *g = gk20a_get_platform(dev)->g; | ||
1043 | struct gk20a_dmabuf_priv *priv; | ||
1044 | static u64 priv_count = 0; | ||
1045 | |||
1046 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
1047 | if (likely(priv)) | ||
1048 | return 0; | ||
1049 | |||
1050 | nvgpu_mutex_acquire(&g->mm.priv_lock); | ||
1051 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
1052 | if (priv) | ||
1053 | goto priv_exist_or_err; | ||
1054 | |||
1055 | priv = nvgpu_kzalloc(g, sizeof(*priv)); | ||
1056 | if (!priv) { | ||
1057 | priv = ERR_PTR(-ENOMEM); | ||
1058 | goto priv_exist_or_err; | ||
1059 | } | ||
1060 | |||
1061 | nvgpu_mutex_init(&priv->lock); | ||
1062 | nvgpu_init_list_node(&priv->states); | ||
1063 | priv->buffer_id = ++priv_count; | ||
1064 | priv->g = g; | ||
1065 | dma_buf_set_drvdata(dmabuf, dev, priv, gk20a_mm_delete_priv); | ||
1066 | |||
1067 | priv_exist_or_err: | ||
1068 | nvgpu_mutex_release(&g->mm.priv_lock); | ||
1069 | if (IS_ERR(priv)) | ||
1070 | return -ENOMEM; | ||
1071 | |||
1072 | return 0; | ||
1073 | } | ||
1074 | |||
1075 | int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g, | ||
1076 | u64 offset, struct gk20a_buffer_state **state) | ||
1077 | { | ||
1078 | int err = 0; | ||
1079 | struct gk20a_dmabuf_priv *priv; | ||
1080 | struct gk20a_buffer_state *s; | ||
1081 | struct device *dev = dev_from_gk20a(g); | ||
1082 | |||
1083 | if (WARN_ON(offset >= (u64)dmabuf->size)) | ||
1084 | return -EINVAL; | ||
1085 | |||
1086 | err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev); | ||
1087 | if (err) | ||
1088 | return err; | ||
1089 | |||
1090 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
1091 | if (WARN_ON(!priv)) | ||
1092 | return -ENOSYS; | ||
1093 | |||
1094 | nvgpu_mutex_acquire(&priv->lock); | ||
1095 | |||
1096 | nvgpu_list_for_each_entry(s, &priv->states, gk20a_buffer_state, list) | ||
1097 | if (s->offset == offset) | ||
1098 | goto out; | ||
1099 | |||
1100 | /* State not found, create state. */ | ||
1101 | s = nvgpu_kzalloc(g, sizeof(*s)); | ||
1102 | if (!s) { | ||
1103 | err = -ENOMEM; | ||
1104 | goto out; | ||
1105 | } | ||
1106 | |||
1107 | s->offset = offset; | ||
1108 | nvgpu_init_list_node(&s->list); | ||
1109 | nvgpu_mutex_init(&s->lock); | ||
1110 | nvgpu_list_add_tail(&s->list, &priv->states); | ||
1111 | |||
1112 | out: | ||
1113 | nvgpu_mutex_release(&priv->lock); | ||
1114 | if (!err) | ||
1115 | *state = s; | ||
1116 | return err; | ||
1117 | |||
1118 | |||
1119 | } | ||
1120 | |||
1121 | int nvgpu_vm_map_buffer(struct vm_gk20a *vm, | 856 | int nvgpu_vm_map_buffer(struct vm_gk20a *vm, |
1122 | int dmabuf_fd, | 857 | int dmabuf_fd, |
1123 | u64 *offset_align, | 858 | u64 *offset_align, |
@@ -1613,34 +1348,3 @@ const struct gk20a_mmu_level *gk20a_mm_get_mmu_levels(struct gk20a *g, | |||
1613 | return (big_page_size == SZ_64K) ? | 1348 | return (big_page_size == SZ_64K) ? |
1614 | gk20a_mm_levels_64k : gk20a_mm_levels_128k; | 1349 | gk20a_mm_levels_64k : gk20a_mm_levels_128k; |
1615 | } | 1350 | } |
1616 | |||
1617 | int gk20a_mm_get_buffer_info(struct device *dev, int dmabuf_fd, | ||
1618 | u64 *buffer_id, u64 *buffer_len) | ||
1619 | { | ||
1620 | struct dma_buf *dmabuf; | ||
1621 | struct gk20a_dmabuf_priv *priv; | ||
1622 | int err = 0; | ||
1623 | |||
1624 | dmabuf = dma_buf_get(dmabuf_fd); | ||
1625 | if (IS_ERR(dmabuf)) { | ||
1626 | dev_warn(dev, "%s: fd %d is not a dmabuf", __func__, dmabuf_fd); | ||
1627 | return PTR_ERR(dmabuf); | ||
1628 | } | ||
1629 | |||
1630 | err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev); | ||
1631 | if (err) { | ||
1632 | dev_warn(dev, "Failed to allocate dmabuf drvdata (err = %d)", | ||
1633 | err); | ||
1634 | goto clean_up; | ||
1635 | } | ||
1636 | |||
1637 | priv = dma_buf_get_drvdata(dmabuf, dev); | ||
1638 | if (likely(priv)) { | ||
1639 | *buffer_id = priv->buffer_id; | ||
1640 | *buffer_len = dmabuf->size; | ||
1641 | } | ||
1642 | |||
1643 | clean_up: | ||
1644 | dma_buf_put(dmabuf); | ||
1645 | return err; | ||
1646 | } | ||
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h index 448496f5..04034d84 100644 --- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h | |||
@@ -117,13 +117,6 @@ gk20a_buffer_state_from_list(struct nvgpu_list_node *node) | |||
117 | ((uintptr_t)node - offsetof(struct gk20a_buffer_state, list)); | 117 | ((uintptr_t)node - offsetof(struct gk20a_buffer_state, list)); |
118 | }; | 118 | }; |
119 | 119 | ||
120 | struct gk20a_comptags { | ||
121 | u32 offset; | ||
122 | u32 lines; | ||
123 | u32 allocated_lines; | ||
124 | bool user_mappable; | ||
125 | }; | ||
126 | |||
127 | struct priv_cmd_queue { | 120 | struct priv_cmd_queue { |
128 | struct nvgpu_mem mem; | 121 | struct nvgpu_mem mem; |
129 | u32 size; /* num of entries in words */ | 122 | u32 size; /* num of entries in words */ |
@@ -369,10 +362,6 @@ void gk20a_locked_gmmu_unmap(struct vm_gk20a *vm, | |||
369 | bool sparse, | 362 | bool sparse, |
370 | struct vm_gk20a_mapping_batch *batch); | 363 | struct vm_gk20a_mapping_batch *batch); |
371 | 364 | ||
372 | struct sg_table *gk20a_mm_pin(struct device *dev, struct dma_buf *dmabuf); | ||
373 | void gk20a_mm_unpin(struct device *dev, struct dma_buf *dmabuf, | ||
374 | struct sg_table *sgt); | ||
375 | |||
376 | /* vm-as interface */ | 365 | /* vm-as interface */ |
377 | struct nvgpu_as_alloc_space_args; | 366 | struct nvgpu_as_alloc_space_args; |
378 | struct nvgpu_as_free_space_args; | 367 | struct nvgpu_as_free_space_args; |
@@ -381,14 +370,6 @@ int gk20a_vm_bind_channel(struct gk20a_as_share *as_share, | |||
381 | struct channel_gk20a *ch); | 370 | struct channel_gk20a *ch); |
382 | int __gk20a_vm_bind_channel(struct vm_gk20a *vm, struct channel_gk20a *ch); | 371 | int __gk20a_vm_bind_channel(struct vm_gk20a *vm, struct channel_gk20a *ch); |
383 | 372 | ||
384 | void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf, | ||
385 | struct gk20a_comptags *comptags); | ||
386 | |||
387 | int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev); | ||
388 | |||
389 | int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g, | ||
390 | u64 offset, struct gk20a_buffer_state **state); | ||
391 | |||
392 | void pde_range_from_vaddr_range(struct vm_gk20a *vm, | 373 | void pde_range_from_vaddr_range(struct vm_gk20a *vm, |
393 | u64 addr_lo, u64 addr_hi, | 374 | u64 addr_lo, u64 addr_hi, |
394 | u32 *pde_lo, u32 *pde_hi); | 375 | u32 *pde_lo, u32 *pde_hi); |
diff --git a/drivers/gpu/nvgpu/gm20b/ltc_gm20b.c b/drivers/gpu/nvgpu/gm20b/ltc_gm20b.c index dee46aaf..84759192 100644 --- a/drivers/gpu/nvgpu/gm20b/ltc_gm20b.c +++ b/drivers/gpu/nvgpu/gm20b/ltc_gm20b.c | |||
@@ -100,7 +100,7 @@ int gm20b_ltc_init_comptags(struct gk20a *g, struct gr_gk20a *gr) | |||
100 | if (err) | 100 | if (err) |
101 | return err; | 101 | return err; |
102 | 102 | ||
103 | err = gk20a_comptag_allocator_init(&gr->comp_tags, max_comptag_lines); | 103 | err = gk20a_comptag_allocator_init(g, &gr->comp_tags, max_comptag_lines); |
104 | if (err) | 104 | if (err) |
105 | return err; | 105 | return err; |
106 | 106 | ||
diff --git a/drivers/gpu/nvgpu/gp10b/ltc_gp10b.c b/drivers/gpu/nvgpu/gp10b/ltc_gp10b.c index 02cab938..9d878402 100644 --- a/drivers/gpu/nvgpu/gp10b/ltc_gp10b.c +++ b/drivers/gpu/nvgpu/gp10b/ltc_gp10b.c | |||
@@ -120,7 +120,7 @@ int gp10b_ltc_init_comptags(struct gk20a *g, struct gr_gk20a *gr) | |||
120 | if (err) | 120 | if (err) |
121 | return err; | 121 | return err; |
122 | 122 | ||
123 | err = gk20a_comptag_allocator_init(&gr->comp_tags, max_comptag_lines); | 123 | err = gk20a_comptag_allocator_init(g, &gr->comp_tags, max_comptag_lines); |
124 | if (err) | 124 | if (err) |
125 | return err; | 125 | return err; |
126 | 126 | ||
diff --git a/drivers/gpu/nvgpu/include/nvgpu/comptags.h b/drivers/gpu/nvgpu/include/nvgpu/comptags.h new file mode 100644 index 00000000..6e3062ec --- /dev/null +++ b/drivers/gpu/nvgpu/include/nvgpu/comptags.h | |||
@@ -0,0 +1,57 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms and conditions of the GNU General Public License, | ||
6 | * version 2, as published by the Free Software Foundation. | ||
7 | * | ||
8 | * This program is distributed in the hope 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 | ||
11 | * more details. | ||
12 | * | ||
13 | * You should have received a copy of the GNU General Public License | ||
14 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
15 | */ | ||
16 | |||
17 | #ifndef __NVGPU_COMPTAGS__ | ||
18 | #define __NVGPU_COMPTAGS__ | ||
19 | |||
20 | #include <nvgpu/lock.h> | ||
21 | |||
22 | struct gk20a; | ||
23 | |||
24 | struct gk20a_comptags { | ||
25 | u32 offset; | ||
26 | u32 lines; | ||
27 | u32 allocated_lines; | ||
28 | bool user_mappable; | ||
29 | }; | ||
30 | |||
31 | struct gk20a_comptag_allocator { | ||
32 | struct gk20a *g; | ||
33 | |||
34 | struct nvgpu_mutex lock; | ||
35 | |||
36 | /* This bitmap starts at ctag 1. 0th cannot be taken. */ | ||
37 | unsigned long *bitmap; | ||
38 | |||
39 | /* Size of bitmap, not max ctags, so one less. */ | ||
40 | unsigned long size; | ||
41 | }; | ||
42 | |||
43 | /* real size here, but first (ctag 0) isn't used */ | ||
44 | int gk20a_comptag_allocator_init(struct gk20a *g, | ||
45 | struct gk20a_comptag_allocator *allocator, | ||
46 | unsigned long size); | ||
47 | void gk20a_comptag_allocator_destroy(struct gk20a *g, | ||
48 | struct gk20a_comptag_allocator *allocator); | ||
49 | |||
50 | int gk20a_comptaglines_alloc(struct gk20a_comptag_allocator *allocator, | ||
51 | u32 *offset, u32 len); | ||
52 | void gk20a_comptaglines_free(struct gk20a_comptag_allocator *allocator, | ||
53 | u32 offset, u32 len); | ||
54 | |||
55 | |||
56 | |||
57 | #endif | ||
diff --git a/drivers/gpu/nvgpu/vgpu/gr_vgpu.c b/drivers/gpu/nvgpu/vgpu/gr_vgpu.c index 96d21c0a..b913847b 100644 --- a/drivers/gpu/nvgpu/vgpu/gr_vgpu.c +++ b/drivers/gpu/nvgpu/vgpu/gr_vgpu.c | |||
@@ -844,7 +844,7 @@ static void vgpu_remove_gr_support(struct gr_gk20a *gr) | |||
844 | { | 844 | { |
845 | gk20a_dbg_fn(""); | 845 | gk20a_dbg_fn(""); |
846 | 846 | ||
847 | gk20a_comptag_allocator_destroy(&gr->comp_tags); | 847 | gk20a_comptag_allocator_destroy(gr->g, &gr->comp_tags); |
848 | 848 | ||
849 | nvgpu_kfree(gr->g, gr->sm_error_states); | 849 | nvgpu_kfree(gr->g, gr->sm_error_states); |
850 | gr->sm_error_states = NULL; | 850 | gr->sm_error_states = NULL; |
diff --git a/drivers/gpu/nvgpu/vgpu/ltc_vgpu.c b/drivers/gpu/nvgpu/vgpu/ltc_vgpu.c index a6848872..fb9558e2 100644 --- a/drivers/gpu/nvgpu/vgpu/ltc_vgpu.c +++ b/drivers/gpu/nvgpu/vgpu/ltc_vgpu.c | |||
@@ -50,7 +50,7 @@ int vgpu_ltc_init_comptags(struct gk20a *g, struct gr_gk20a *gr) | |||
50 | if (max_comptag_lines < 2) | 50 | if (max_comptag_lines < 2) |
51 | return -ENXIO; | 51 | return -ENXIO; |
52 | 52 | ||
53 | err = gk20a_comptag_allocator_init(&gr->comp_tags, max_comptag_lines); | 53 | err = gk20a_comptag_allocator_init(g, &gr->comp_tags, max_comptag_lines); |
54 | if (err) | 54 | if (err) |
55 | return err; | 55 | return err; |
56 | 56 | ||