| 1 |
/* |
| 2 |
* Copyright 2023 Red Hat Inc. |
| 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 18 |
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 19 |
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 20 |
* OTHER DEALINGS IN THE SOFTWARE. |
| 21 |
*/ |
| 22 |
#include <rm/gr.h> |
| 23 |
|
| 24 |
#include <core/memory.h> |
| 25 |
#include <subdev/gsp.h> |
| 26 |
#include <subdev/mmu/vmm.h> |
| 27 |
#include <engine/fifo/priv.h> |
| 28 |
#include <engine/gr/priv.h> |
| 29 |
|
| 30 |
#include <nvif/if900d.h> |
| 31 |
|
| 32 |
#include <nvhw/drf.h> |
| 33 |
|
| 34 |
#include "nvrm/gr.h" |
| 35 |
#include "nvrm/vmm.h" |
| 36 |
|
| 37 |
#define r535_gr(p) container_of((p), struct r535_gr, base) |
| 38 |
|
| 39 |
static void * |
| 40 |
r535_gr_chan_dtor(struct nvkm_object *object) |
| 41 |
{ |
| 42 |
struct r535_gr_chan *grc = container_of(object, typeof(*grc), object); |
| 43 |
struct r535_gr *gr = grc->gr; |
| 44 |
|
| 45 |
for (int i = 0; i < gr->ctxbuf_nr; i++) { |
| 46 |
nvkm_vmm_put(grc->vmm, &grc->vma[i]); |
| 47 |
nvkm_memory_unref(&grc->mem[i]); |
| 48 |
} |
| 49 |
|
| 50 |
nvkm_vmm_unref(&grc->vmm); |
| 51 |
return grc; |
| 52 |
} |
| 53 |
|
| 54 |
static const struct nvkm_object_func |
| 55 |
r535_gr_chan = { |
| 56 |
.dtor = r535_gr_chan_dtor, |
| 57 |
}; |
| 58 |
|
| 59 |
int |
| 60 |
r535_gr_promote_ctx(struct r535_gr *gr, bool golden, struct nvkm_vmm *vmm, |
| 61 |
struct nvkm_memory **pmem, struct nvkm_vma **pvma, |
| 62 |
struct nvkm_gsp_object *chan) |
| 63 |
{ |
| 64 |
struct nvkm_subdev *subdev = &gr->base.engine.subdev; |
| 65 |
struct nvkm_device *device = subdev->device; |
| 66 |
NV2080_CTRL_GPU_PROMOTE_CTX_PARAMS *ctrl; |
| 67 |
|
| 68 |
ctrl = nvkm_gsp_rm_ctrl_get(&vmm->rm.device.subdevice, |
| 69 |
NV2080_CTRL_CMD_GPU_PROMOTE_CTX, sizeof(*ctrl)); |
| 70 |
if (WARN_ON(IS_ERR(ctrl))) |
| 71 |
return PTR_ERR(ctrl); |
| 72 |
|
| 73 |
ctrl->engineType = 1; |
| 74 |
ctrl->hChanClient = vmm->rm.client.object.handle; |
| 75 |
ctrl->hObject = chan->handle; |
| 76 |
|
| 77 |
for (int i = 0; i < gr->ctxbuf_nr; i++) { |
| 78 |
NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ENTRY *entry = |
| 79 |
&ctrl->promoteEntry[ctrl->entryCount]; |
| 80 |
const bool alloc = golden || !gr->ctxbuf[i].global; |
| 81 |
int ret; |
| 82 |
|
| 83 |
entry->bufferId = gr->ctxbuf[i].bufferId; |
| 84 |
entry->bInitialize = gr->ctxbuf[i].init && alloc; |
| 85 |
|
| 86 |
if (alloc) { |
| 87 |
ret = nvkm_memory_new(device, gr->ctxbuf[i].init ? |
| 88 |
NVKM_MEM_TARGET_INST : NVKM_MEM_TARGET_INST_SR_LOST, |
| 89 |
gr->ctxbuf[i].size, 1 << gr->ctxbuf[i].page, |
| 90 |
gr->ctxbuf[i].init, &pmem[i]); |
| 91 |
if (WARN_ON(ret)) |
| 92 |
return ret; |
| 93 |
|
| 94 |
if (gr->ctxbuf[i].bufferId == |
| 95 |
NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_PRIV_ACCESS_MAP) |
| 96 |
entry->bNonmapped = 1; |
| 97 |
} else { |
| 98 |
if (gr->ctxbuf[i].bufferId == |
| 99 |
NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_UNRESTRICTED_PRIV_ACCESS_MAP) |
| 100 |
continue; |
| 101 |
|
| 102 |
pmem[i] = nvkm_memory_ref(gr->ctxbuf_mem[i]); |
| 103 |
} |
| 104 |
|
| 105 |
if (!entry->bNonmapped) { |
| 106 |
struct gf100_vmm_map_v0 args = { |
| 107 |
.priv = 1, |
| 108 |
.ro = gr->ctxbuf[i].ro, |
| 109 |
}; |
| 110 |
|
| 111 |
mutex_lock(&vmm->mutex.vmm); |
| 112 |
ret = nvkm_vmm_get_locked(vmm, false, true, false, 0, gr->ctxbuf[i].align, |
| 113 |
nvkm_memory_size(pmem[i]), &pvma[i]); |
| 114 |
mutex_unlock(&vmm->mutex.vmm); |
| 115 |
if (ret) |
| 116 |
return ret; |
| 117 |
|
| 118 |
ret = nvkm_memory_map(pmem[i], 0, vmm, pvma[i], &args, sizeof(args)); |
| 119 |
if (ret) |
| 120 |
return ret; |
| 121 |
|
| 122 |
entry->gpuVirtAddr = pvma[i]->addr; |
| 123 |
} |
| 124 |
|
| 125 |
if (entry->bInitialize) { |
| 126 |
entry->gpuPhysAddr = nvkm_memory_addr(pmem[i]); |
| 127 |
entry->size = gr->ctxbuf[i].size; |
| 128 |
entry->physAttr = 4; |
| 129 |
} |
| 130 |
|
| 131 |
nvkm_debug(subdev, |
| 132 |
"promote %02d: pa %016llx/%08x sz %016llx va %016llx init:%d nm:%d\n", |
| 133 |
entry->bufferId, entry->gpuPhysAddr, entry->physAttr, entry->size, |
| 134 |
entry->gpuVirtAddr, entry->bInitialize, entry->bNonmapped); |
| 135 |
|
| 136 |
ctrl->entryCount++; |
| 137 |
} |
| 138 |
|
| 139 |
return nvkm_gsp_rm_ctrl_wr(&vmm->rm.device.subdevice, ctrl); |
| 140 |
} |
| 141 |
|
| 142 |
int |
| 143 |
r535_gr_chan_new(struct nvkm_gr *base, struct nvkm_chan *chan, const struct nvkm_oclass *oclass, |
| 144 |
struct nvkm_object **pobject) |
| 145 |
{ |
| 146 |
struct r535_gr *gr = r535_gr(base); |
| 147 |
struct r535_gr_chan *grc; |
| 148 |
int ret; |
| 149 |
|
| 150 |
if (!(grc = kzalloc_obj(*grc))) |
| 151 |
return -ENOMEM; |
| 152 |
|
| 153 |
nvkm_object_ctor(&r535_gr_chan, oclass, &grc->object); |
| 154 |
grc->gr = gr; |
| 155 |
grc->vmm = nvkm_vmm_ref(chan->vmm); |
| 156 |
grc->chan = chan; |
| 157 |
*pobject = &grc->object; |
| 158 |
|
| 159 |
ret = r535_gr_promote_ctx(gr, false, grc->vmm, grc->mem, grc->vma, &chan->rm.object); |
| 160 |
if (ret) |
| 161 |
return ret; |
| 162 |
|
| 163 |
return 0; |
| 164 |
} |
| 165 |
|
| 166 |
u64 |
| 167 |
r535_gr_units(struct nvkm_gr *gr) |
| 168 |
{ |
| 169 |
struct nvkm_gsp *gsp = gr->engine.subdev.device->gsp; |
| 170 |
|
| 171 |
return (gsp->gr.tpcs << 8) | gsp->gr.gpcs; |
| 172 |
} |
| 173 |
|
| 174 |
void |
| 175 |
r535_gr_get_ctxbuf_info(struct r535_gr *gr, int i, |
| 176 |
struct NV2080_CTRL_INTERNAL_ENGINE_CONTEXT_BUFFER_INFO *info) |
| 177 |
{ |
| 178 |
struct nvkm_subdev *subdev = &gr->base.engine.subdev; |
| 179 |
static const struct { |
| 180 |
u32 id0; /* NV0080_CTRL_FIFO_GET_ENGINE_CONTEXT_PROPERTIES_ENGINE_ID */ |
| 181 |
u32 id1; /* NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID */ |
| 182 |
bool global; |
| 183 |
bool init; |
| 184 |
bool ro; |
| 185 |
} map[] = { |
| 186 |
#define _A(n,N,G,I,R) { .id0 = NV0080_CTRL_FIFO_GET_ENGINE_CONTEXT_PROPERTIES_ENGINE_ID_##n, \ |
| 187 |
.id1 = NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_##N, \ |
| 188 |
.global = (G), .init = (I), .ro = (R) } |
| 189 |
#define _B(N,G,I,R) _A(GRAPHICS_##N, N, (G), (I), (R)) |
| 190 |
/* global init ro */ |
| 191 |
_A( GRAPHICS, MAIN, false, true, false), |
| 192 |
_B( PATCH, false, true, false), |
| 193 |
_A( GRAPHICS_BUNDLE_CB, BUFFER_BUNDLE_CB, true, false, false), |
| 194 |
_B( PAGEPOOL, true, false, false), |
| 195 |
_B( ATTRIBUTE_CB, true, false, false), |
| 196 |
_B( RTV_CB_GLOBAL, true, false, false), |
| 197 |
_B( FECS_EVENT, true, true, false), |
| 198 |
_B( PRIV_ACCESS_MAP, true, true, true), |
| 199 |
#undef _B |
| 200 |
#undef _A |
| 201 |
}; |
| 202 |
u32 size = info->size; |
| 203 |
u8 align, page; |
| 204 |
int id; |
| 205 |
|
| 206 |
for (id = 0; id < ARRAY_SIZE(map); id++) { |
| 207 |
if (map[id].id0 == i) |
| 208 |
break; |
| 209 |
} |
| 210 |
|
| 211 |
nvkm_debug(subdev, "%02x: size:0x%08x %s\n", i, |
| 212 |
size, (id < ARRAY_SIZE(map)) ? "*" : ""); |
| 213 |
if (id >= ARRAY_SIZE(map)) |
| 214 |
return; |
| 215 |
|
| 216 |
if (map[id].id1 == NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_MAIN) |
| 217 |
size = ALIGN(size, 0x1000) + 64 * 0x1000; /* per-subctx headers */ |
| 218 |
|
| 219 |
if (size >= 1 << 21) page = 21; |
| 220 |
else if (size >= 1 << 16) page = 16; |
| 221 |
else page = 12; |
| 222 |
|
| 223 |
if (map[id].id1 == NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_ATTRIBUTE_CB) |
| 224 |
align = order_base_2(size); |
| 225 |
else |
| 226 |
align = page; |
| 227 |
|
| 228 |
if (WARN_ON(gr->ctxbuf_nr == ARRAY_SIZE(gr->ctxbuf))) |
| 229 |
return; |
| 230 |
|
| 231 |
gr->ctxbuf[gr->ctxbuf_nr].bufferId = map[id].id1; |
| 232 |
gr->ctxbuf[gr->ctxbuf_nr].size = size; |
| 233 |
gr->ctxbuf[gr->ctxbuf_nr].page = page; |
| 234 |
gr->ctxbuf[gr->ctxbuf_nr].align = align; |
| 235 |
gr->ctxbuf[gr->ctxbuf_nr].global = map[id].global; |
| 236 |
gr->ctxbuf[gr->ctxbuf_nr].init = map[id].init; |
| 237 |
gr->ctxbuf[gr->ctxbuf_nr].ro = map[id].ro; |
| 238 |
gr->ctxbuf_nr++; |
| 239 |
|
| 240 |
if (map[id].id1 == NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_PRIV_ACCESS_MAP) { |
| 241 |
if (WARN_ON(gr->ctxbuf_nr == ARRAY_SIZE(gr->ctxbuf))) |
| 242 |
return; |
| 243 |
|
| 244 |
gr->ctxbuf[gr->ctxbuf_nr] = gr->ctxbuf[gr->ctxbuf_nr - 1]; |
| 245 |
gr->ctxbuf[gr->ctxbuf_nr].bufferId = |
| 246 |
NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_UNRESTRICTED_PRIV_ACCESS_MAP; |
| 247 |
gr->ctxbuf_nr++; |
| 248 |
} |
| 249 |
} |
| 250 |
|
| 251 |
static int |
| 252 |
r535_gr_get_ctxbufs_and_zcull_info(struct r535_gr *gr) |
| 253 |
{ |
| 254 |
NV2080_CTRL_INTERNAL_STATIC_GR_GET_CONTEXT_BUFFERS_INFO_PARAMS *info; |
| 255 |
struct nvkm_subdev *subdev = &gr->base.engine.subdev; |
| 256 |
struct nvkm_gsp *gsp = subdev->device->gsp; |
| 257 |
|
| 258 |
info = nvkm_gsp_rm_ctrl_rd(&gsp->internal.device.subdevice, |
| 259 |
NV2080_CTRL_CMD_INTERNAL_STATIC_KGR_GET_CONTEXT_BUFFERS_INFO, |
| 260 |
sizeof(*info)); |
| 261 |
if (WARN_ON(IS_ERR(info))) |
| 262 |
return PTR_ERR(info); |
| 263 |
|
| 264 |
for (int i = 0; i < ARRAY_SIZE(info->engineContextBuffersInfo[0].engine); i++) |
| 265 |
r535_gr_get_ctxbuf_info(gr, i, &info->engineContextBuffersInfo[0].engine[i]); |
| 266 |
|
| 267 |
nvkm_gsp_rm_ctrl_done(&gsp->internal.device.subdevice, info); |
| 268 |
|
| 269 |
gr->base.has_zcull_info = false; |
| 270 |
|
| 271 |
return 0; |
| 272 |
} |
| 273 |
|
| 274 |
int |
| 275 |
r535_gr_oneinit(struct nvkm_gr *base) |
| 276 |
{ |
| 277 |
struct r535_gr *gr = container_of(base, typeof(*gr), base); |
| 278 |
struct nvkm_subdev *subdev = &gr->base.engine.subdev; |
| 279 |
struct nvkm_device *device = subdev->device; |
| 280 |
struct nvkm_gsp *gsp = device->gsp; |
| 281 |
struct nvkm_rm *rm = gsp->rm; |
| 282 |
struct { |
| 283 |
struct nvkm_memory *inst; |
| 284 |
struct nvkm_vmm *vmm; |
| 285 |
struct nvkm_gsp_object chan; |
| 286 |
struct nvkm_vma *vma[R515_GR_MAX_CTXBUFS]; |
| 287 |
} golden = {}; |
| 288 |
struct nvkm_gsp_object threed; |
| 289 |
int ret; |
| 290 |
|
| 291 |
/* Allocate a channel to use for golden context init. */ |
| 292 |
ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x12000, 0, true, &golden.inst); |
| 293 |
if (ret) |
| 294 |
goto done; |
| 295 |
|
| 296 |
ret = nvkm_vmm_new(device, 0x1000, 0, NULL, 0, NULL, "grGoldenVmm", &golden.vmm); |
| 297 |
if (ret) |
| 298 |
goto done; |
| 299 |
|
| 300 |
ret = r535_mmu_vaspace_new(golden.vmm, NVKM_RM_VASPACE, false); |
| 301 |
if (ret) |
| 302 |
goto done; |
| 303 |
|
| 304 |
ret = rm->api->fifo->chan.alloc(&golden.vmm->rm.device, NVKM_RM_CHAN(0), |
| 305 |
1, 0, true, rm->api->fifo->rsvd_chids, |
| 306 |
nvkm_memory_addr(golden.inst), |
| 307 |
nvkm_memory_addr(golden.inst) + 0x1000, |
| 308 |
nvkm_memory_addr(golden.inst) + 0x2000, |
| 309 |
golden.vmm, 0, 0x1000, &golden.chan); |
| 310 |
if (ret) |
| 311 |
goto done; |
| 312 |
|
| 313 |
/* Fetch context buffer info from RM and allocate each of them here to use |
| 314 |
* during golden context init (or later as a global context buffer). |
| 315 |
* |
| 316 |
* Also build the information that'll be used to create channel contexts. |
| 317 |
*/ |
| 318 |
ret = rm->api->gr->get_ctxbufs_and_zcull_info(gr); |
| 319 |
if (ret) |
| 320 |
goto done; |
| 321 |
|
| 322 |
/* Promote golden context to RM. */ |
| 323 |
ret = r535_gr_promote_ctx(gr, true, golden.vmm, gr->ctxbuf_mem, golden.vma, &golden.chan); |
| 324 |
if (ret) |
| 325 |
goto done; |
| 326 |
|
| 327 |
/* Allocate 3D class on channel to trigger golden context init in RM. */ |
| 328 |
ret = nvkm_gsp_rm_alloc(&golden.chan, NVKM_RM_THREED, rm->gpu->gr.class.threed, 0, &threed); |
| 329 |
if (ret) |
| 330 |
goto done; |
| 331 |
|
| 332 |
/* There's no need to keep the golden channel around, as RM caches the context. */ |
| 333 |
nvkm_gsp_rm_free(&threed); |
| 334 |
done: |
| 335 |
nvkm_gsp_rm_free(&golden.chan); |
| 336 |
for (int i = gr->ctxbuf_nr - 1; i >= 0; i--) |
| 337 |
nvkm_vmm_put(golden.vmm, &golden.vma[i]); |
| 338 |
nvkm_vmm_unref(&golden.vmm); |
| 339 |
nvkm_memory_unref(&golden.inst); |
| 340 |
return ret; |
| 341 |
|
| 342 |
} |
| 343 |
|
| 344 |
void * |
| 345 |
r535_gr_dtor(struct nvkm_gr *base) |
| 346 |
{ |
| 347 |
struct r535_gr *gr = r535_gr(base); |
| 348 |
|
| 349 |
while (gr->ctxbuf_nr) |
| 350 |
nvkm_memory_unref(&gr->ctxbuf_mem[--gr->ctxbuf_nr]); |
| 351 |
|
| 352 |
kfree(gr->base.func); |
| 353 |
return gr; |
| 354 |
} |
| 355 |
|
| 356 |
const struct nvkm_rm_api_gr |
| 357 |
r535_gr = { |
| 358 |
.get_ctxbufs_and_zcull_info = r535_gr_get_ctxbufs_and_zcull_info, |
| 359 |
}; |