diff options
author | Alex Waterman <alexw@nvidia.com> | 2017-07-31 15:32:07 -0400 |
---|---|---|
committer | mobile promotions <svcmobile_promotions@nvidia.com> | 2017-10-10 11:01:04 -0400 |
commit | 3c37701377459fbea2b460e1b9c65a863dfb04b2 (patch) | |
tree | 60a57a260dc8bcbd81089af8e9e966074e3bc881 /drivers/gpu/nvgpu/common | |
parent | b61306795b53647e8d1d4279417df1e1fe0f4b86 (diff) |
gpu: nvgpu: Split VIDMEM support from mm_gk20a.c
Split VIDMEM support into its own code files organized as such:
common/mm/vidmem.c - Base vidmem support
common/linux/vidmem.c - Linux specific user-space interaction
include/nvgpu/vidmem.h - Vidmem API definitions
Also use the config to enable/disable VIDMEM support in the makefile
and remove as many CONFIG_GK20A_VIDMEM preprocessor checks as possible
from the source code.
And lastly update a while-loop that iterated over an SGT to use the
new for_each construct for iterating over SGTs.
Currently this organization is not perfectly adhered to. More patches
will fix that.
JIRA NVGPU-30
JIRA NVGPU-138
Change-Id: Ic0f4d2cf38b65849c7dc350a69b175421477069c
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1540705
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu/common')
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/dma.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/ioctl_ctrl.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/nvgpu_mem.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/vidmem.c | 268 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/vm.c | 29 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/mm/gmmu.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/mm/nvgpu_mem.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/mm/vidmem.c | 259 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/common/pramin.c | 1 |
9 files changed, 562 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/common/linux/dma.c b/drivers/gpu/nvgpu/common/linux/dma.c index 7b892731..2ed1cc5a 100644 --- a/drivers/gpu/nvgpu/common/linux/dma.c +++ b/drivers/gpu/nvgpu/common/linux/dma.c | |||
@@ -25,6 +25,7 @@ | |||
25 | #include <nvgpu/gmmu.h> | 25 | #include <nvgpu/gmmu.h> |
26 | #include <nvgpu/kmem.h> | 26 | #include <nvgpu/kmem.h> |
27 | #include <nvgpu/enabled.h> | 27 | #include <nvgpu/enabled.h> |
28 | #include <nvgpu/vidmem.h> | ||
28 | 29 | ||
29 | #include <nvgpu/linux/dma.h> | 30 | #include <nvgpu/linux/dma.h> |
30 | 31 | ||
diff --git a/drivers/gpu/nvgpu/common/linux/ioctl_ctrl.c b/drivers/gpu/nvgpu/common/linux/ioctl_ctrl.c index 220cb3e7..87ae0a3d 100644 --- a/drivers/gpu/nvgpu/common/linux/ioctl_ctrl.c +++ b/drivers/gpu/nvgpu/common/linux/ioctl_ctrl.c | |||
@@ -25,6 +25,7 @@ | |||
25 | #include <nvgpu/kmem.h> | 25 | #include <nvgpu/kmem.h> |
26 | #include <nvgpu/bug.h> | 26 | #include <nvgpu/bug.h> |
27 | #include <nvgpu/bus.h> | 27 | #include <nvgpu/bus.h> |
28 | #include <nvgpu/vidmem.h> | ||
28 | 29 | ||
29 | #include "ioctl_ctrl.h" | 30 | #include "ioctl_ctrl.h" |
30 | #include "ioctl_tsg.h" | 31 | #include "ioctl_tsg.h" |
diff --git a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c index 0be41a44..e8aea0be 100644 --- a/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c +++ b/drivers/gpu/nvgpu/common/linux/nvgpu_mem.c | |||
@@ -22,6 +22,7 @@ | |||
22 | #include <nvgpu/bug.h> | 22 | #include <nvgpu/bug.h> |
23 | #include <nvgpu/enabled.h> | 23 | #include <nvgpu/enabled.h> |
24 | #include <nvgpu/kmem.h> | 24 | #include <nvgpu/kmem.h> |
25 | #include <nvgpu/vidmem.h> | ||
25 | 26 | ||
26 | #include <nvgpu/linux/dma.h> | 27 | #include <nvgpu/linux/dma.h> |
27 | 28 | ||
diff --git a/drivers/gpu/nvgpu/common/linux/vidmem.c b/drivers/gpu/nvgpu/common/linux/vidmem.c new file mode 100644 index 00000000..e89dd07a --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/vidmem.c | |||
@@ -0,0 +1,268 @@ | |||
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 <linux/platform/tegra/tegra_fd.h> | ||
20 | |||
21 | #include <nvgpu/dma.h> | ||
22 | #include <nvgpu/enabled.h> | ||
23 | #include <nvgpu/vidmem.h> | ||
24 | #include <nvgpu/nvgpu_mem.h> | ||
25 | #include <nvgpu/page_allocator.h> | ||
26 | |||
27 | #include <nvgpu/linux/dma.h> | ||
28 | |||
29 | #include "gk20a/gk20a.h" | ||
30 | #include "gk20a/mm_gk20a.h" | ||
31 | |||
32 | #include "vm_priv.h" | ||
33 | |||
34 | void set_vidmem_page_alloc(struct scatterlist *sgl, u64 addr) | ||
35 | { | ||
36 | /* set bit 0 to indicate vidmem allocation */ | ||
37 | sg_dma_address(sgl) = (addr | 1ULL); | ||
38 | } | ||
39 | |||
40 | bool is_vidmem_page_alloc(u64 addr) | ||
41 | { | ||
42 | return !!(addr & 1ULL); | ||
43 | } | ||
44 | |||
45 | struct nvgpu_page_alloc *get_vidmem_page_alloc(struct scatterlist *sgl) | ||
46 | { | ||
47 | u64 addr; | ||
48 | |||
49 | addr = sg_dma_address(sgl); | ||
50 | |||
51 | if (is_vidmem_page_alloc(addr)) | ||
52 | addr = addr & ~1ULL; | ||
53 | else | ||
54 | WARN_ON(1); | ||
55 | |||
56 | return (struct nvgpu_page_alloc *)(uintptr_t)addr; | ||
57 | } | ||
58 | |||
59 | static struct sg_table *gk20a_vidbuf_map_dma_buf( | ||
60 | struct dma_buf_attachment *attach, enum dma_data_direction dir) | ||
61 | { | ||
62 | struct gk20a_vidmem_buf *buf = attach->dmabuf->priv; | ||
63 | |||
64 | return buf->mem->priv.sgt; | ||
65 | } | ||
66 | |||
67 | static void gk20a_vidbuf_unmap_dma_buf(struct dma_buf_attachment *attach, | ||
68 | struct sg_table *sgt, | ||
69 | enum dma_data_direction dir) | ||
70 | { | ||
71 | } | ||
72 | |||
73 | static void gk20a_vidbuf_release(struct dma_buf *dmabuf) | ||
74 | { | ||
75 | struct gk20a_vidmem_buf *buf = dmabuf->priv; | ||
76 | |||
77 | gk20a_dbg_fn(""); | ||
78 | |||
79 | if (buf->dmabuf_priv) | ||
80 | buf->dmabuf_priv_delete(buf->dmabuf_priv); | ||
81 | |||
82 | nvgpu_dma_free(buf->g, buf->mem); | ||
83 | nvgpu_kfree(buf->g, buf); | ||
84 | } | ||
85 | |||
86 | static void *gk20a_vidbuf_kmap(struct dma_buf *dmabuf, unsigned long page_num) | ||
87 | { | ||
88 | WARN_ON("Not supported"); | ||
89 | return NULL; | ||
90 | } | ||
91 | |||
92 | static void *gk20a_vidbuf_kmap_atomic(struct dma_buf *dmabuf, | ||
93 | unsigned long page_num) | ||
94 | { | ||
95 | WARN_ON("Not supported"); | ||
96 | return NULL; | ||
97 | } | ||
98 | |||
99 | static int gk20a_vidbuf_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma) | ||
100 | { | ||
101 | return -EINVAL; | ||
102 | } | ||
103 | |||
104 | static int gk20a_vidbuf_set_private(struct dma_buf *dmabuf, | ||
105 | struct device *dev, void *priv, void (*delete)(void *priv)) | ||
106 | { | ||
107 | struct gk20a_vidmem_buf *buf = dmabuf->priv; | ||
108 | |||
109 | buf->dmabuf_priv = priv; | ||
110 | buf->dmabuf_priv_delete = delete; | ||
111 | |||
112 | return 0; | ||
113 | } | ||
114 | |||
115 | static void *gk20a_vidbuf_get_private(struct dma_buf *dmabuf, | ||
116 | struct device *dev) | ||
117 | { | ||
118 | struct gk20a_vidmem_buf *buf = dmabuf->priv; | ||
119 | |||
120 | return buf->dmabuf_priv; | ||
121 | } | ||
122 | |||
123 | static const struct dma_buf_ops gk20a_vidbuf_ops = { | ||
124 | .map_dma_buf = gk20a_vidbuf_map_dma_buf, | ||
125 | .unmap_dma_buf = gk20a_vidbuf_unmap_dma_buf, | ||
126 | .release = gk20a_vidbuf_release, | ||
127 | .kmap_atomic = gk20a_vidbuf_kmap_atomic, | ||
128 | .kmap = gk20a_vidbuf_kmap, | ||
129 | .mmap = gk20a_vidbuf_mmap, | ||
130 | .set_drvdata = gk20a_vidbuf_set_private, | ||
131 | .get_drvdata = gk20a_vidbuf_get_private, | ||
132 | }; | ||
133 | |||
134 | static struct dma_buf *gk20a_vidbuf_export(struct gk20a_vidmem_buf *buf) | ||
135 | { | ||
136 | DEFINE_DMA_BUF_EXPORT_INFO(exp_info); | ||
137 | |||
138 | exp_info.priv = buf; | ||
139 | exp_info.ops = &gk20a_vidbuf_ops; | ||
140 | exp_info.size = buf->mem->size; | ||
141 | exp_info.flags = O_RDWR; | ||
142 | |||
143 | return dma_buf_export(&exp_info); | ||
144 | } | ||
145 | |||
146 | struct gk20a *gk20a_vidmem_buf_owner(struct dma_buf *dmabuf) | ||
147 | { | ||
148 | struct gk20a_vidmem_buf *buf = dmabuf->priv; | ||
149 | |||
150 | if (dmabuf->ops != &gk20a_vidbuf_ops) | ||
151 | return NULL; | ||
152 | |||
153 | return buf->g; | ||
154 | } | ||
155 | |||
156 | int gk20a_vidmem_buf_alloc(struct gk20a *g, size_t bytes) | ||
157 | { | ||
158 | struct gk20a_vidmem_buf *buf; | ||
159 | int err = 0, fd; | ||
160 | |||
161 | gk20a_dbg_fn(""); | ||
162 | |||
163 | buf = nvgpu_kzalloc(g, sizeof(*buf)); | ||
164 | if (!buf) | ||
165 | return -ENOMEM; | ||
166 | |||
167 | buf->g = g; | ||
168 | |||
169 | if (!g->mm.vidmem.cleared) { | ||
170 | nvgpu_mutex_acquire(&g->mm.vidmem.first_clear_mutex); | ||
171 | if (!g->mm.vidmem.cleared) { | ||
172 | err = gk20a_vidmem_clear_all(g); | ||
173 | if (err) { | ||
174 | nvgpu_err(g, | ||
175 | "failed to clear whole vidmem"); | ||
176 | goto err_kfree; | ||
177 | } | ||
178 | } | ||
179 | nvgpu_mutex_release(&g->mm.vidmem.first_clear_mutex); | ||
180 | } | ||
181 | |||
182 | buf->mem = nvgpu_kzalloc(g, sizeof(struct nvgpu_mem)); | ||
183 | if (!buf->mem) | ||
184 | goto err_kfree; | ||
185 | |||
186 | buf->mem->mem_flags |= NVGPU_MEM_FLAG_USER_MEM; | ||
187 | |||
188 | err = nvgpu_dma_alloc_vid(g, bytes, buf->mem); | ||
189 | if (err) | ||
190 | goto err_memfree; | ||
191 | |||
192 | buf->dmabuf = gk20a_vidbuf_export(buf); | ||
193 | if (IS_ERR(buf->dmabuf)) { | ||
194 | err = PTR_ERR(buf->dmabuf); | ||
195 | goto err_bfree; | ||
196 | } | ||
197 | |||
198 | fd = tegra_alloc_fd(current->files, 1024, O_RDWR); | ||
199 | if (fd < 0) { | ||
200 | /* ->release frees what we have done */ | ||
201 | dma_buf_put(buf->dmabuf); | ||
202 | return fd; | ||
203 | } | ||
204 | |||
205 | /* fclose() on this drops one ref, freeing the dma buf */ | ||
206 | fd_install(fd, buf->dmabuf->file); | ||
207 | |||
208 | return fd; | ||
209 | |||
210 | err_bfree: | ||
211 | nvgpu_dma_free(g, buf->mem); | ||
212 | err_memfree: | ||
213 | nvgpu_kfree(g, buf->mem); | ||
214 | err_kfree: | ||
215 | nvgpu_kfree(g, buf); | ||
216 | return err; | ||
217 | } | ||
218 | |||
219 | int gk20a_vidbuf_access_memory(struct gk20a *g, struct dma_buf *dmabuf, | ||
220 | void *buffer, u64 offset, u64 size, u32 cmd) | ||
221 | { | ||
222 | struct gk20a_vidmem_buf *vidmem_buf; | ||
223 | struct nvgpu_mem *mem; | ||
224 | int err = 0; | ||
225 | |||
226 | if (gk20a_dmabuf_aperture(g, dmabuf) != APERTURE_VIDMEM) | ||
227 | return -EINVAL; | ||
228 | |||
229 | vidmem_buf = dmabuf->priv; | ||
230 | mem = vidmem_buf->mem; | ||
231 | |||
232 | switch (cmd) { | ||
233 | case NVGPU_DBG_GPU_IOCTL_ACCESS_FB_MEMORY_CMD_READ: | ||
234 | nvgpu_mem_rd_n(g, mem, offset, buffer, size); | ||
235 | break; | ||
236 | |||
237 | case NVGPU_DBG_GPU_IOCTL_ACCESS_FB_MEMORY_CMD_WRITE: | ||
238 | nvgpu_mem_wr_n(g, mem, offset, buffer, size); | ||
239 | break; | ||
240 | |||
241 | default: | ||
242 | err = -EINVAL; | ||
243 | } | ||
244 | |||
245 | return err; | ||
246 | } | ||
247 | |||
248 | void gk20a_vidmem_clear_mem_worker(struct work_struct *work) | ||
249 | { | ||
250 | struct mm_gk20a *mm = container_of(work, struct mm_gk20a, | ||
251 | vidmem.clear_mem_worker); | ||
252 | struct gk20a *g = mm->g; | ||
253 | struct nvgpu_mem *mem; | ||
254 | |||
255 | while ((mem = get_pending_mem_desc(mm)) != NULL) { | ||
256 | gk20a_gmmu_clear_vidmem_mem(g, mem); | ||
257 | nvgpu_free(mem->allocator, | ||
258 | (u64)get_vidmem_page_alloc(mem->priv.sgt->sgl)); | ||
259 | nvgpu_free_sgtable(g, &mem->priv.sgt); | ||
260 | |||
261 | WARN_ON(nvgpu_atomic64_sub_return(mem->aligned_size, | ||
262 | &g->mm.vidmem.bytes_pending) < 0); | ||
263 | mem->size = 0; | ||
264 | mem->aperture = APERTURE_INVALID; | ||
265 | |||
266 | nvgpu_kfree(g, mem); | ||
267 | } | ||
268 | } | ||
diff --git a/drivers/gpu/nvgpu/common/linux/vm.c b/drivers/gpu/nvgpu/common/linux/vm.c index 2e29f0f7..f4ac3d41 100644 --- a/drivers/gpu/nvgpu/common/linux/vm.c +++ b/drivers/gpu/nvgpu/common/linux/vm.c | |||
@@ -23,6 +23,8 @@ | |||
23 | #include <nvgpu/vm_area.h> | 23 | #include <nvgpu/vm_area.h> |
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> | ||
27 | #include <nvgpu/enabled.h> | ||
26 | 28 | ||
27 | #include <nvgpu/linux/nvgpu_mem.h> | 29 | #include <nvgpu/linux/nvgpu_mem.h> |
28 | 30 | ||
@@ -34,6 +36,33 @@ | |||
34 | #include "vm_priv.h" | 36 | #include "vm_priv.h" |
35 | #include "os_linux.h" | 37 | #include "os_linux.h" |
36 | 38 | ||
39 | /* | ||
40 | * Temporary location for this code until a dmabuf.c file exists. | ||
41 | */ | ||
42 | enum nvgpu_aperture gk20a_dmabuf_aperture(struct gk20a *g, | ||
43 | struct dma_buf *dmabuf) | ||
44 | { | ||
45 | struct gk20a *buf_owner = gk20a_vidmem_buf_owner(dmabuf); | ||
46 | bool unified_memory = nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY); | ||
47 | |||
48 | if (buf_owner == NULL) { | ||
49 | /* Not nvgpu-allocated, assume system memory */ | ||
50 | return APERTURE_SYSMEM; | ||
51 | } else if (WARN_ON(buf_owner == g && unified_memory)) { | ||
52 | /* Looks like our video memory, but this gpu doesn't support | ||
53 | * it. Warn about a bug and bail out */ | ||
54 | nvgpu_warn(g, | ||
55 | "dmabuf is our vidmem but we don't have local vidmem"); | ||
56 | return APERTURE_INVALID; | ||
57 | } else if (buf_owner != g) { | ||
58 | /* Someone else's vidmem */ | ||
59 | return APERTURE_INVALID; | ||
60 | } else { | ||
61 | /* Yay, buf_owner == g */ | ||
62 | return APERTURE_VIDMEM; | ||
63 | } | ||
64 | } | ||
65 | |||
37 | static struct nvgpu_mapped_buf *__nvgpu_vm_find_mapped_buf_reverse( | 66 | static struct nvgpu_mapped_buf *__nvgpu_vm_find_mapped_buf_reverse( |
38 | struct vm_gk20a *vm, struct dma_buf *dmabuf, u32 kind) | 67 | struct vm_gk20a *vm, struct dma_buf *dmabuf, u32 kind) |
39 | { | 68 | { |
diff --git a/drivers/gpu/nvgpu/common/mm/gmmu.c b/drivers/gpu/nvgpu/common/mm/gmmu.c index f61ec3fc..1eed3a3b 100644 --- a/drivers/gpu/nvgpu/common/mm/gmmu.c +++ b/drivers/gpu/nvgpu/common/mm/gmmu.c | |||
@@ -28,6 +28,7 @@ | |||
28 | #include <nvgpu/enabled.h> | 28 | #include <nvgpu/enabled.h> |
29 | #include <nvgpu/page_allocator.h> | 29 | #include <nvgpu/page_allocator.h> |
30 | #include <nvgpu/barrier.h> | 30 | #include <nvgpu/barrier.h> |
31 | #include <nvgpu/vidmem.h> | ||
31 | 32 | ||
32 | #include "gk20a/gk20a.h" | 33 | #include "gk20a/gk20a.h" |
33 | #include "gk20a/mm_gk20a.h" | 34 | #include "gk20a/mm_gk20a.h" |
diff --git a/drivers/gpu/nvgpu/common/mm/nvgpu_mem.c b/drivers/gpu/nvgpu/common/mm/nvgpu_mem.c index faee482d..2b6e6e6a 100644 --- a/drivers/gpu/nvgpu/common/mm/nvgpu_mem.c +++ b/drivers/gpu/nvgpu/common/mm/nvgpu_mem.c | |||
@@ -23,6 +23,7 @@ | |||
23 | #include <nvgpu/kmem.h> | 23 | #include <nvgpu/kmem.h> |
24 | #include <nvgpu/nvgpu_mem.h> | 24 | #include <nvgpu/nvgpu_mem.h> |
25 | #include <nvgpu/dma.h> | 25 | #include <nvgpu/dma.h> |
26 | #include <nvgpu/vidmem.h> | ||
26 | 27 | ||
27 | #include "gk20a/gk20a.h" | 28 | #include "gk20a/gk20a.h" |
28 | 29 | ||
diff --git a/drivers/gpu/nvgpu/common/mm/vidmem.c b/drivers/gpu/nvgpu/common/mm/vidmem.c new file mode 100644 index 00000000..1ba07ca6 --- /dev/null +++ b/drivers/gpu/nvgpu/common/mm/vidmem.c | |||
@@ -0,0 +1,259 @@ | |||
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 <linux/scatterlist.h> | ||
24 | |||
25 | #include <nvgpu/vidmem.h> | ||
26 | #include <nvgpu/page_allocator.h> | ||
27 | |||
28 | #include "gk20a/gk20a.h" | ||
29 | #include "gk20a/mm_gk20a.h" | ||
30 | |||
31 | void gk20a_vidmem_destroy(struct gk20a *g) | ||
32 | { | ||
33 | if (nvgpu_alloc_initialized(&g->mm.vidmem.allocator)) | ||
34 | nvgpu_alloc_destroy(&g->mm.vidmem.allocator); | ||
35 | } | ||
36 | |||
37 | int gk20a_vidmem_clear_all(struct gk20a *g) | ||
38 | { | ||
39 | struct mm_gk20a *mm = &g->mm; | ||
40 | struct gk20a_fence *gk20a_fence_out = NULL; | ||
41 | u64 region2_base = 0; | ||
42 | int err = 0; | ||
43 | |||
44 | if (mm->vidmem.ce_ctx_id == (u32)~0) | ||
45 | return -EINVAL; | ||
46 | |||
47 | err = gk20a_ce_execute_ops(g, | ||
48 | mm->vidmem.ce_ctx_id, | ||
49 | 0, | ||
50 | mm->vidmem.base, | ||
51 | mm->vidmem.bootstrap_base - mm->vidmem.base, | ||
52 | 0x00000000, | ||
53 | NVGPU_CE_DST_LOCATION_LOCAL_FB, | ||
54 | NVGPU_CE_MEMSET, | ||
55 | NULL, | ||
56 | 0, | ||
57 | NULL); | ||
58 | if (err) { | ||
59 | nvgpu_err(g, | ||
60 | "Failed to clear vidmem region 1 : %d", err); | ||
61 | return err; | ||
62 | } | ||
63 | |||
64 | region2_base = mm->vidmem.bootstrap_base + mm->vidmem.bootstrap_size; | ||
65 | |||
66 | err = gk20a_ce_execute_ops(g, | ||
67 | mm->vidmem.ce_ctx_id, | ||
68 | 0, | ||
69 | region2_base, | ||
70 | mm->vidmem.size - region2_base, | ||
71 | 0x00000000, | ||
72 | NVGPU_CE_DST_LOCATION_LOCAL_FB, | ||
73 | NVGPU_CE_MEMSET, | ||
74 | NULL, | ||
75 | 0, | ||
76 | &gk20a_fence_out); | ||
77 | if (err) { | ||
78 | nvgpu_err(g, | ||
79 | "Failed to clear vidmem region 2 : %d", err); | ||
80 | return err; | ||
81 | } | ||
82 | |||
83 | if (gk20a_fence_out) { | ||
84 | struct nvgpu_timeout timeout; | ||
85 | |||
86 | nvgpu_timeout_init(g, &timeout, | ||
87 | gk20a_get_gr_idle_timeout(g), | ||
88 | NVGPU_TIMER_CPU_TIMER); | ||
89 | |||
90 | do { | ||
91 | err = gk20a_fence_wait(g, gk20a_fence_out, | ||
92 | gk20a_get_gr_idle_timeout(g)); | ||
93 | } while (err == -ERESTARTSYS && | ||
94 | !nvgpu_timeout_expired(&timeout)); | ||
95 | |||
96 | gk20a_fence_put(gk20a_fence_out); | ||
97 | if (err) { | ||
98 | nvgpu_err(g, | ||
99 | "fence wait failed for CE execute ops"); | ||
100 | return err; | ||
101 | } | ||
102 | } | ||
103 | |||
104 | mm->vidmem.cleared = true; | ||
105 | |||
106 | return 0; | ||
107 | } | ||
108 | |||
109 | int gk20a_init_vidmem(struct mm_gk20a *mm) | ||
110 | { | ||
111 | struct gk20a *g = mm->g; | ||
112 | size_t size = g->ops.mm.get_vidmem_size ? | ||
113 | g->ops.mm.get_vidmem_size(g) : 0; | ||
114 | u64 bootstrap_base, bootstrap_size, base; | ||
115 | u64 default_page_size = SZ_64K; | ||
116 | int err; | ||
117 | |||
118 | static struct nvgpu_alloc_carveout wpr_co = | ||
119 | NVGPU_CARVEOUT("wpr-region", 0, SZ_16M); | ||
120 | |||
121 | if (!size) | ||
122 | return 0; | ||
123 | |||
124 | wpr_co.base = size - SZ_256M; | ||
125 | bootstrap_base = wpr_co.base; | ||
126 | bootstrap_size = SZ_16M; | ||
127 | base = default_page_size; | ||
128 | |||
129 | /* | ||
130 | * Bootstrap allocator for use before the CE is initialized (CE | ||
131 | * initialization requires vidmem but we want to use the CE to zero | ||
132 | * out vidmem before allocating it... | ||
133 | */ | ||
134 | err = nvgpu_page_allocator_init(g, &g->mm.vidmem.bootstrap_allocator, | ||
135 | "vidmem-bootstrap", | ||
136 | bootstrap_base, bootstrap_size, | ||
137 | SZ_4K, 0); | ||
138 | |||
139 | err = nvgpu_page_allocator_init(g, &g->mm.vidmem.allocator, | ||
140 | "vidmem", | ||
141 | base, size - base, | ||
142 | default_page_size, | ||
143 | GPU_ALLOC_4K_VIDMEM_PAGES); | ||
144 | if (err) { | ||
145 | nvgpu_err(g, "Failed to register vidmem for size %zu: %d", | ||
146 | size, err); | ||
147 | return err; | ||
148 | } | ||
149 | |||
150 | /* Reserve bootstrap region in vidmem allocator */ | ||
151 | nvgpu_alloc_reserve_carveout(&g->mm.vidmem.allocator, &wpr_co); | ||
152 | |||
153 | mm->vidmem.base = base; | ||
154 | mm->vidmem.size = size - base; | ||
155 | mm->vidmem.bootstrap_base = bootstrap_base; | ||
156 | mm->vidmem.bootstrap_size = bootstrap_size; | ||
157 | |||
158 | nvgpu_mutex_init(&mm->vidmem.first_clear_mutex); | ||
159 | |||
160 | INIT_WORK(&mm->vidmem.clear_mem_worker, gk20a_vidmem_clear_mem_worker); | ||
161 | nvgpu_atomic64_set(&mm->vidmem.bytes_pending, 0); | ||
162 | nvgpu_init_list_node(&mm->vidmem.clear_list_head); | ||
163 | nvgpu_mutex_init(&mm->vidmem.clear_list_mutex); | ||
164 | |||
165 | gk20a_dbg_info("registered vidmem: %zu MB", size / SZ_1M); | ||
166 | |||
167 | return 0; | ||
168 | } | ||
169 | |||
170 | int gk20a_vidmem_get_space(struct gk20a *g, u64 *space) | ||
171 | { | ||
172 | struct nvgpu_allocator *allocator = &g->mm.vidmem.allocator; | ||
173 | |||
174 | gk20a_dbg_fn(""); | ||
175 | |||
176 | if (!nvgpu_alloc_initialized(allocator)) | ||
177 | return -ENOSYS; | ||
178 | |||
179 | nvgpu_mutex_acquire(&g->mm.vidmem.clear_list_mutex); | ||
180 | *space = nvgpu_alloc_space(allocator) + | ||
181 | nvgpu_atomic64_read(&g->mm.vidmem.bytes_pending); | ||
182 | nvgpu_mutex_release(&g->mm.vidmem.clear_list_mutex); | ||
183 | return 0; | ||
184 | } | ||
185 | |||
186 | int gk20a_gmmu_clear_vidmem_mem(struct gk20a *g, struct nvgpu_mem *mem) | ||
187 | { | ||
188 | struct gk20a_fence *gk20a_fence_out = NULL; | ||
189 | struct gk20a_fence *gk20a_last_fence = NULL; | ||
190 | struct nvgpu_page_alloc *alloc = NULL; | ||
191 | void *sgl = NULL; | ||
192 | int err = 0; | ||
193 | |||
194 | if (g->mm.vidmem.ce_ctx_id == (u32)~0) | ||
195 | return -EINVAL; | ||
196 | |||
197 | alloc = get_vidmem_page_alloc(mem->priv.sgt->sgl); | ||
198 | |||
199 | nvgpu_sgt_for_each_sgl(sgl, &alloc->sgt) { | ||
200 | if (gk20a_last_fence) | ||
201 | gk20a_fence_put(gk20a_last_fence); | ||
202 | |||
203 | err = gk20a_ce_execute_ops(g, | ||
204 | g->mm.vidmem.ce_ctx_id, | ||
205 | 0, | ||
206 | nvgpu_sgt_get_phys(&alloc->sgt, sgl), | ||
207 | nvgpu_sgt_get_length(&alloc->sgt, sgl), | ||
208 | 0x00000000, | ||
209 | NVGPU_CE_DST_LOCATION_LOCAL_FB, | ||
210 | NVGPU_CE_MEMSET, | ||
211 | NULL, | ||
212 | 0, | ||
213 | &gk20a_fence_out); | ||
214 | |||
215 | if (err) { | ||
216 | nvgpu_err(g, | ||
217 | "Failed gk20a_ce_execute_ops[%d]", err); | ||
218 | return err; | ||
219 | } | ||
220 | |||
221 | gk20a_last_fence = gk20a_fence_out; | ||
222 | } | ||
223 | |||
224 | if (gk20a_last_fence) { | ||
225 | struct nvgpu_timeout timeout; | ||
226 | |||
227 | nvgpu_timeout_init(g, &timeout, | ||
228 | gk20a_get_gr_idle_timeout(g), | ||
229 | NVGPU_TIMER_CPU_TIMER); | ||
230 | |||
231 | do { | ||
232 | err = gk20a_fence_wait(g, gk20a_last_fence, | ||
233 | gk20a_get_gr_idle_timeout(g)); | ||
234 | } while (err == -ERESTARTSYS && | ||
235 | !nvgpu_timeout_expired(&timeout)); | ||
236 | |||
237 | gk20a_fence_put(gk20a_last_fence); | ||
238 | if (err) | ||
239 | nvgpu_err(g, | ||
240 | "fence wait failed for CE execute ops"); | ||
241 | } | ||
242 | |||
243 | return err; | ||
244 | } | ||
245 | |||
246 | struct nvgpu_mem *get_pending_mem_desc(struct mm_gk20a *mm) | ||
247 | { | ||
248 | struct nvgpu_mem *mem = NULL; | ||
249 | |||
250 | nvgpu_mutex_acquire(&mm->vidmem.clear_list_mutex); | ||
251 | if (!nvgpu_list_empty(&mm->vidmem.clear_list_head)) { | ||
252 | mem = nvgpu_list_first_entry(&mm->vidmem.clear_list_head, | ||
253 | nvgpu_mem, clear_list_entry); | ||
254 | nvgpu_list_del(&mem->clear_list_entry); | ||
255 | } | ||
256 | nvgpu_mutex_release(&mm->vidmem.clear_list_mutex); | ||
257 | |||
258 | return mem; | ||
259 | } | ||
diff --git a/drivers/gpu/nvgpu/common/pramin.c b/drivers/gpu/nvgpu/common/pramin.c index 4f7d6248..56179a6b 100644 --- a/drivers/gpu/nvgpu/common/pramin.c +++ b/drivers/gpu/nvgpu/common/pramin.c | |||
@@ -23,6 +23,7 @@ | |||
23 | #include <nvgpu/pramin.h> | 23 | #include <nvgpu/pramin.h> |
24 | #include <nvgpu/page_allocator.h> | 24 | #include <nvgpu/page_allocator.h> |
25 | #include <nvgpu/enabled.h> | 25 | #include <nvgpu/enabled.h> |
26 | #include <nvgpu/vidmem.h> | ||
26 | 27 | ||
27 | #include "gk20a/gk20a.h" | 28 | #include "gk20a/gk20a.h" |
28 | 29 | ||