diff options
Diffstat (limited to 'drivers/gpu/nvgpu/common/linux/vgpu/gp10b/vgpu_gr_gp10b.c')
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/vgpu/gp10b/vgpu_gr_gp10b.c | 332 |
1 files changed, 332 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/common/linux/vgpu/gp10b/vgpu_gr_gp10b.c b/drivers/gpu/nvgpu/common/linux/vgpu/gp10b/vgpu_gr_gp10b.c new file mode 100644 index 00000000..efc9c595 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/vgpu/gp10b/vgpu_gr_gp10b.c | |||
@@ -0,0 +1,332 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2015-2017, 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 <nvgpu/kmem.h> | ||
18 | #include <nvgpu/dma.h> | ||
19 | #include <nvgpu/bug.h> | ||
20 | |||
21 | #include "common/linux/vgpu/vgpu.h" | ||
22 | #include "common/linux/vgpu/gm20b/vgpu_gr_gm20b.h" | ||
23 | |||
24 | #include "vgpu_gr_gp10b.h" | ||
25 | |||
26 | #include <nvgpu/hw/gp10b/hw_gr_gp10b.h> | ||
27 | |||
28 | void vgpu_gr_gp10b_free_gr_ctx(struct gk20a *g, struct vm_gk20a *vm, | ||
29 | struct gr_ctx_desc *gr_ctx) | ||
30 | { | ||
31 | struct tegra_vgpu_cmd_msg msg = {0}; | ||
32 | struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; | ||
33 | int err; | ||
34 | |||
35 | gk20a_dbg_fn(""); | ||
36 | |||
37 | if (!gr_ctx || !gr_ctx->mem.gpu_va) | ||
38 | return; | ||
39 | |||
40 | msg.cmd = TEGRA_VGPU_CMD_GR_CTX_FREE; | ||
41 | msg.handle = vgpu_get_handle(g); | ||
42 | p->gr_ctx_handle = gr_ctx->virt_ctx; | ||
43 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
44 | WARN_ON(err || msg.ret); | ||
45 | |||
46 | __nvgpu_vm_free_va(vm, gr_ctx->mem.gpu_va, gmmu_page_size_kernel); | ||
47 | |||
48 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.pagepool_ctxsw_buffer); | ||
49 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.betacb_ctxsw_buffer); | ||
50 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.spill_ctxsw_buffer); | ||
51 | nvgpu_dma_unmap_free(vm, &gr_ctx->t18x.preempt_ctxsw_buffer); | ||
52 | |||
53 | nvgpu_kfree(g, gr_ctx); | ||
54 | } | ||
55 | |||
56 | int vgpu_gr_gp10b_alloc_gr_ctx(struct gk20a *g, | ||
57 | struct gr_ctx_desc **__gr_ctx, | ||
58 | struct vm_gk20a *vm, | ||
59 | u32 class, | ||
60 | u32 flags) | ||
61 | { | ||
62 | struct gr_ctx_desc *gr_ctx; | ||
63 | u32 graphics_preempt_mode = 0; | ||
64 | u32 compute_preempt_mode = 0; | ||
65 | struct vgpu_priv_data *priv = vgpu_get_priv_data(g); | ||
66 | int err; | ||
67 | |||
68 | gk20a_dbg_fn(""); | ||
69 | |||
70 | err = vgpu_gr_alloc_gr_ctx(g, __gr_ctx, vm, class, flags); | ||
71 | if (err) | ||
72 | return err; | ||
73 | |||
74 | gr_ctx = *__gr_ctx; | ||
75 | |||
76 | if (flags & NVGPU_OBJ_CTX_FLAGS_SUPPORT_GFXP) | ||
77 | graphics_preempt_mode = NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
78 | if (flags & NVGPU_OBJ_CTX_FLAGS_SUPPORT_CILP) | ||
79 | compute_preempt_mode = NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
80 | |||
81 | if (priv->constants.force_preempt_mode && !graphics_preempt_mode && | ||
82 | !compute_preempt_mode) { | ||
83 | graphics_preempt_mode = g->ops.gr.is_valid_gfx_class(g, class) ? | ||
84 | NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP : 0; | ||
85 | compute_preempt_mode = | ||
86 | g->ops.gr.is_valid_compute_class(g, class) ? | ||
87 | NVGPU_PREEMPTION_MODE_COMPUTE_CTA : 0; | ||
88 | } | ||
89 | |||
90 | if (graphics_preempt_mode || compute_preempt_mode) { | ||
91 | if (g->ops.gr.set_ctxsw_preemption_mode) { | ||
92 | err = g->ops.gr.set_ctxsw_preemption_mode(g, gr_ctx, vm, | ||
93 | class, graphics_preempt_mode, compute_preempt_mode); | ||
94 | if (err) { | ||
95 | nvgpu_err(g, | ||
96 | "set_ctxsw_preemption_mode failed"); | ||
97 | goto fail; | ||
98 | } | ||
99 | } else { | ||
100 | err = -ENOSYS; | ||
101 | goto fail; | ||
102 | } | ||
103 | } | ||
104 | |||
105 | gk20a_dbg_fn("done"); | ||
106 | return err; | ||
107 | |||
108 | fail: | ||
109 | vgpu_gr_gp10b_free_gr_ctx(g, vm, gr_ctx); | ||
110 | return err; | ||
111 | } | ||
112 | |||
113 | int vgpu_gr_gp10b_set_ctxsw_preemption_mode(struct gk20a *g, | ||
114 | struct gr_ctx_desc *gr_ctx, | ||
115 | struct vm_gk20a *vm, u32 class, | ||
116 | u32 graphics_preempt_mode, | ||
117 | u32 compute_preempt_mode) | ||
118 | { | ||
119 | struct tegra_vgpu_cmd_msg msg = {}; | ||
120 | struct tegra_vgpu_gr_bind_ctxsw_buffers_params *p = | ||
121 | &msg.params.gr_bind_ctxsw_buffers; | ||
122 | int err = 0; | ||
123 | |||
124 | if (g->ops.gr.is_valid_gfx_class(g, class) && | ||
125 | g->gr.t18x.ctx_vars.force_preemption_gfxp) | ||
126 | graphics_preempt_mode = NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
127 | |||
128 | if (g->ops.gr.is_valid_compute_class(g, class) && | ||
129 | g->gr.t18x.ctx_vars.force_preemption_cilp) | ||
130 | compute_preempt_mode = NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
131 | |||
132 | /* check for invalid combinations */ | ||
133 | if ((graphics_preempt_mode == 0) && (compute_preempt_mode == 0)) | ||
134 | return -EINVAL; | ||
135 | |||
136 | if ((graphics_preempt_mode == NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP) && | ||
137 | (compute_preempt_mode == NVGPU_PREEMPTION_MODE_COMPUTE_CILP)) | ||
138 | return -EINVAL; | ||
139 | |||
140 | /* set preemption modes */ | ||
141 | switch (graphics_preempt_mode) { | ||
142 | case NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP: | ||
143 | { | ||
144 | u32 spill_size = | ||
145 | gr_gpc0_swdx_rm_spill_buffer_size_256b_default_v() * | ||
146 | gr_gpc0_swdx_rm_spill_buffer_size_256b_byte_granularity_v(); | ||
147 | u32 pagepool_size = g->ops.gr.pagepool_default_size(g) * | ||
148 | gr_scc_pagepool_total_pages_byte_granularity_v(); | ||
149 | u32 betacb_size = g->gr.attrib_cb_default_size + | ||
150 | (gr_gpc0_ppc0_cbm_beta_cb_size_v_gfxp_v() - | ||
151 | gr_gpc0_ppc0_cbm_beta_cb_size_v_default_v()); | ||
152 | u32 attrib_cb_size = (betacb_size + g->gr.alpha_cb_size) * | ||
153 | gr_gpc0_ppc0_cbm_beta_cb_size_v_granularity_v() * | ||
154 | g->gr.max_tpc_count; | ||
155 | struct nvgpu_mem *desc; | ||
156 | |||
157 | attrib_cb_size = ALIGN(attrib_cb_size, 128); | ||
158 | |||
159 | gk20a_dbg_info("gfxp context preempt size=%d", | ||
160 | g->gr.t18x.ctx_vars.preempt_image_size); | ||
161 | gk20a_dbg_info("gfxp context spill size=%d", spill_size); | ||
162 | gk20a_dbg_info("gfxp context pagepool size=%d", pagepool_size); | ||
163 | gk20a_dbg_info("gfxp context attrib cb size=%d", | ||
164 | attrib_cb_size); | ||
165 | |||
166 | err = gr_gp10b_alloc_buffer(vm, | ||
167 | g->gr.t18x.ctx_vars.preempt_image_size, | ||
168 | &gr_ctx->t18x.preempt_ctxsw_buffer); | ||
169 | if (err) { | ||
170 | err = -ENOMEM; | ||
171 | goto fail; | ||
172 | } | ||
173 | desc = &gr_ctx->t18x.preempt_ctxsw_buffer; | ||
174 | p->gpu_va[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAIN] = desc->gpu_va; | ||
175 | p->size[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAIN] = desc->size; | ||
176 | |||
177 | err = gr_gp10b_alloc_buffer(vm, | ||
178 | spill_size, | ||
179 | &gr_ctx->t18x.spill_ctxsw_buffer); | ||
180 | if (err) { | ||
181 | err = -ENOMEM; | ||
182 | goto fail; | ||
183 | } | ||
184 | desc = &gr_ctx->t18x.spill_ctxsw_buffer; | ||
185 | p->gpu_va[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_SPILL] = desc->gpu_va; | ||
186 | p->size[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_SPILL] = desc->size; | ||
187 | |||
188 | err = gr_gp10b_alloc_buffer(vm, | ||
189 | pagepool_size, | ||
190 | &gr_ctx->t18x.pagepool_ctxsw_buffer); | ||
191 | if (err) { | ||
192 | err = -ENOMEM; | ||
193 | goto fail; | ||
194 | } | ||
195 | desc = &gr_ctx->t18x.pagepool_ctxsw_buffer; | ||
196 | p->gpu_va[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_PAGEPOOL] = | ||
197 | desc->gpu_va; | ||
198 | p->size[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_PAGEPOOL] = desc->size; | ||
199 | |||
200 | err = gr_gp10b_alloc_buffer(vm, | ||
201 | attrib_cb_size, | ||
202 | &gr_ctx->t18x.betacb_ctxsw_buffer); | ||
203 | if (err) { | ||
204 | err = -ENOMEM; | ||
205 | goto fail; | ||
206 | } | ||
207 | desc = &gr_ctx->t18x.betacb_ctxsw_buffer; | ||
208 | p->gpu_va[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_BETACB] = | ||
209 | desc->gpu_va; | ||
210 | p->size[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_BETACB] = desc->size; | ||
211 | |||
212 | gr_ctx->graphics_preempt_mode = NVGPU_PREEMPTION_MODE_GRAPHICS_GFXP; | ||
213 | p->mode = TEGRA_VGPU_GR_CTXSW_PREEMPTION_MODE_GFX_GFXP; | ||
214 | break; | ||
215 | } | ||
216 | case NVGPU_PREEMPTION_MODE_GRAPHICS_WFI: | ||
217 | gr_ctx->graphics_preempt_mode = graphics_preempt_mode; | ||
218 | break; | ||
219 | |||
220 | default: | ||
221 | break; | ||
222 | } | ||
223 | |||
224 | if (g->ops.gr.is_valid_compute_class(g, class)) { | ||
225 | switch (compute_preempt_mode) { | ||
226 | case NVGPU_PREEMPTION_MODE_COMPUTE_WFI: | ||
227 | gr_ctx->compute_preempt_mode = | ||
228 | NVGPU_PREEMPTION_MODE_COMPUTE_WFI; | ||
229 | p->mode = TEGRA_VGPU_GR_CTXSW_PREEMPTION_MODE_WFI; | ||
230 | break; | ||
231 | case NVGPU_PREEMPTION_MODE_COMPUTE_CTA: | ||
232 | gr_ctx->compute_preempt_mode = | ||
233 | NVGPU_PREEMPTION_MODE_COMPUTE_CTA; | ||
234 | p->mode = | ||
235 | TEGRA_VGPU_GR_CTXSW_PREEMPTION_MODE_COMPUTE_CTA; | ||
236 | break; | ||
237 | case NVGPU_PREEMPTION_MODE_COMPUTE_CILP: | ||
238 | gr_ctx->compute_preempt_mode = | ||
239 | NVGPU_PREEMPTION_MODE_COMPUTE_CILP; | ||
240 | p->mode = | ||
241 | TEGRA_VGPU_GR_CTXSW_PREEMPTION_MODE_COMPUTE_CILP; | ||
242 | break; | ||
243 | default: | ||
244 | break; | ||
245 | } | ||
246 | } | ||
247 | |||
248 | if (gr_ctx->graphics_preempt_mode || gr_ctx->compute_preempt_mode) { | ||
249 | msg.cmd = TEGRA_VGPU_CMD_CHANNEL_BIND_GR_CTXSW_BUFFERS; | ||
250 | msg.handle = vgpu_get_handle(g); | ||
251 | p->gr_ctx_handle = gr_ctx->virt_ctx; | ||
252 | err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); | ||
253 | if (err || msg.ret) { | ||
254 | err = -ENOMEM; | ||
255 | goto fail; | ||
256 | } | ||
257 | } | ||
258 | |||
259 | return err; | ||
260 | |||
261 | fail: | ||
262 | nvgpu_err(g, "%s failed %d", __func__, err); | ||
263 | return err; | ||
264 | } | ||
265 | |||
266 | int vgpu_gr_gp10b_set_preemption_mode(struct channel_gk20a *ch, | ||
267 | u32 graphics_preempt_mode, | ||
268 | u32 compute_preempt_mode) | ||
269 | { | ||
270 | struct gr_ctx_desc *gr_ctx = ch->ch_ctx.gr_ctx; | ||
271 | struct gk20a *g = ch->g; | ||
272 | struct tsg_gk20a *tsg; | ||
273 | struct vm_gk20a *vm; | ||
274 | u32 class; | ||
275 | int err; | ||
276 | |||
277 | class = ch->obj_class; | ||
278 | if (!class) | ||
279 | return -EINVAL; | ||
280 | |||
281 | /* skip setting anything if both modes are already set */ | ||
282 | if (graphics_preempt_mode && | ||
283 | (graphics_preempt_mode == gr_ctx->graphics_preempt_mode)) | ||
284 | graphics_preempt_mode = 0; | ||
285 | |||
286 | if (compute_preempt_mode && | ||
287 | (compute_preempt_mode == gr_ctx->compute_preempt_mode)) | ||
288 | compute_preempt_mode = 0; | ||
289 | |||
290 | if (graphics_preempt_mode == 0 && compute_preempt_mode == 0) | ||
291 | return 0; | ||
292 | |||
293 | if (gk20a_is_channel_marked_as_tsg(ch)) { | ||
294 | tsg = &g->fifo.tsg[ch->tsgid]; | ||
295 | vm = tsg->vm; | ||
296 | } else { | ||
297 | vm = ch->vm; | ||
298 | } | ||
299 | |||
300 | if (g->ops.gr.set_ctxsw_preemption_mode) { | ||
301 | err = g->ops.gr.set_ctxsw_preemption_mode(g, gr_ctx, vm, class, | ||
302 | graphics_preempt_mode, | ||
303 | compute_preempt_mode); | ||
304 | if (err) { | ||
305 | nvgpu_err(g, "set_ctxsw_preemption_mode failed"); | ||
306 | return err; | ||
307 | } | ||
308 | } else { | ||
309 | err = -ENOSYS; | ||
310 | } | ||
311 | |||
312 | return err; | ||
313 | } | ||
314 | |||
315 | int vgpu_gr_gp10b_init_ctx_state(struct gk20a *g) | ||
316 | { | ||
317 | struct vgpu_priv_data *priv = vgpu_get_priv_data(g); | ||
318 | int err; | ||
319 | |||
320 | gk20a_dbg_fn(""); | ||
321 | |||
322 | err = vgpu_gr_init_ctx_state(g); | ||
323 | if (err) | ||
324 | return err; | ||
325 | |||
326 | g->gr.t18x.ctx_vars.preempt_image_size = | ||
327 | priv->constants.preempt_ctx_size; | ||
328 | if (!g->gr.t18x.ctx_vars.preempt_image_size) | ||
329 | return -EINVAL; | ||
330 | |||
331 | return 0; | ||
332 | } | ||