| 1 |
/* |
| 2 |
* Copyright 2014 Advanced Micro Devices, 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 |
*/ |
| 23 |
|
| 24 |
/** |
| 25 |
* \file amdgpu_device.c |
| 26 |
* |
| 27 |
* Implementation of functions for AMD GPU device |
| 28 |
* |
| 29 |
*/ |
| 30 |
|
| 31 |
#include <sys/stat.h> |
| 32 |
#include <errno.h> |
| 33 |
#include <string.h> |
| 34 |
#include <stdio.h> |
| 35 |
#include <stdlib.h> |
| 36 |
#include <unistd.h> |
| 37 |
#include <fcntl.h> |
| 38 |
|
| 39 |
#include "xf86drm.h" |
| 40 |
#include "amdgpu_drm.h" |
| 41 |
#include "amdgpu_internal.h" |
| 42 |
#include "util_math.h" |
| 43 |
|
| 44 |
#define PTR_TO_UINT(x) ((unsigned)((intptr_t)(x))) |
| 45 |
|
| 46 |
static pthread_mutex_t dev_mutex = PTHREAD_MUTEX_INITIALIZER; |
| 47 |
static amdgpu_device_handle dev_list; |
| 48 |
|
| 49 |
static int fd_compare(int fd1, int fd2) |
| 50 |
{ |
| 51 |
char *name1 = drmGetPrimaryDeviceNameFromFd(fd1); |
| 52 |
char *name2 = drmGetPrimaryDeviceNameFromFd(fd2); |
| 53 |
int result; |
| 54 |
|
| 55 |
if (name1 == NULL || name2 == NULL) { |
| 56 |
free(name1); |
| 57 |
free(name2); |
| 58 |
return 0; |
| 59 |
} |
| 60 |
|
| 61 |
result = strcmp(name1, name2); |
| 62 |
free(name1); |
| 63 |
free(name2); |
| 64 |
|
| 65 |
return result; |
| 66 |
} |
| 67 |
|
| 68 |
/** |
| 69 |
* Get the authenticated form fd, |
| 70 |
* |
| 71 |
* \param fd - \c [in] File descriptor for AMD GPU device |
| 72 |
* \param auth - \c [out] Pointer to output the fd is authenticated or not |
| 73 |
* A render node fd, output auth = 0 |
| 74 |
* A legacy fd, get the authenticated for compatibility root |
| 75 |
* |
| 76 |
* \return 0 on success\n |
| 77 |
* >0 - AMD specific error code\n |
| 78 |
* <0 - Negative POSIX Error code |
| 79 |
*/ |
| 80 |
static int amdgpu_get_auth(int fd, int *auth) |
| 81 |
{ |
| 82 |
int r = 0; |
| 83 |
drm_client_t client = {}; |
| 84 |
|
| 85 |
if (drmGetNodeTypeFromFd(fd) == DRM_NODE_RENDER) |
| 86 |
*auth = 0; |
| 87 |
else { |
| 88 |
client.idx = 0; |
| 89 |
r = drmIoctl(fd, DRM_IOCTL_GET_CLIENT, &client); |
| 90 |
if (!r) |
| 91 |
*auth = client.auth; |
| 92 |
} |
| 93 |
return r; |
| 94 |
} |
| 95 |
|
| 96 |
static void amdgpu_device_free_internal(amdgpu_device_handle dev) |
| 97 |
{ |
| 98 |
/* Remove dev from dev_list, if it was added there. */ |
| 99 |
if (dev == dev_list) { |
| 100 |
dev_list = dev->next; |
| 101 |
} else { |
| 102 |
for (amdgpu_device_handle node = dev_list; node; node = node->next) { |
| 103 |
if (node->next == dev) { |
| 104 |
node->next = dev->next; |
| 105 |
break; |
| 106 |
} |
| 107 |
} |
| 108 |
} |
| 109 |
|
| 110 |
close(dev->fd); |
| 111 |
if ((dev->flink_fd >= 0) && (dev->fd != dev->flink_fd)) |
| 112 |
close(dev->flink_fd); |
| 113 |
|
| 114 |
amdgpu_vamgr_deinit(&dev->va_mgr.vamgr_32); |
| 115 |
amdgpu_vamgr_deinit(&dev->va_mgr.vamgr_low); |
| 116 |
amdgpu_vamgr_deinit(&dev->va_mgr.vamgr_high_32); |
| 117 |
amdgpu_vamgr_deinit(&dev->va_mgr.vamgr_high); |
| 118 |
handle_table_fini(&dev->bo_handles); |
| 119 |
handle_table_fini(&dev->bo_flink_names); |
| 120 |
pthread_mutex_destroy(&dev->bo_table_mutex); |
| 121 |
free(dev->marketing_name); |
| 122 |
free(dev); |
| 123 |
} |
| 124 |
|
| 125 |
/** |
| 126 |
* Assignment between two amdgpu_device pointers with reference counting. |
| 127 |
* |
| 128 |
* Usage: |
| 129 |
* struct amdgpu_device *dst = ... , *src = ...; |
| 130 |
* |
| 131 |
* dst = src; |
| 132 |
* // No reference counting. Only use this when you need to move |
| 133 |
* // a reference from one pointer to another. |
| 134 |
* |
| 135 |
* amdgpu_device_reference(&dst, src); |
| 136 |
* // Reference counters are updated. dst is decremented and src is |
| 137 |
* // incremented. dst is freed if its reference counter is 0. |
| 138 |
*/ |
| 139 |
static void amdgpu_device_reference(struct amdgpu_device **dst, |
| 140 |
struct amdgpu_device *src) |
| 141 |
{ |
| 142 |
if (update_references(&(*dst)->refcount, &src->refcount)) |
| 143 |
amdgpu_device_free_internal(*dst); |
| 144 |
*dst = src; |
| 145 |
} |
| 146 |
|
| 147 |
static int amdgpu_query_gfx_level_major(amdgpu_device_handle dev, |
| 148 |
uint8_t *gfx_ip_version_major) |
| 149 |
{ |
| 150 |
struct drm_amdgpu_info_hw_ip ip_info; |
| 151 |
uint32_t gfx_ip_count = 0; |
| 152 |
int r; |
| 153 |
|
| 154 |
*gfx_ip_version_major = 0; |
| 155 |
|
| 156 |
r = amdgpu_query_hw_ip_count(dev, AMDGPU_HW_IP_GFX, &gfx_ip_count); |
| 157 |
if (r) |
| 158 |
return r; |
| 159 |
|
| 160 |
/* No graphics support. */ |
| 161 |
if (gfx_ip_count == 0) |
| 162 |
return 0; |
| 163 |
|
| 164 |
memset(&ip_info, 0, sizeof(ip_info)); |
| 165 |
|
| 166 |
r = amdgpu_query_hw_ip_info(dev, AMDGPU_HW_IP_GFX, 0, &ip_info); |
| 167 |
if (r) |
| 168 |
return r; |
| 169 |
|
| 170 |
/* GFX6-8 don't set ip_discovery_version. */ |
| 171 |
if (dev->minor_version >= 48 && ip_info.ip_discovery_version) { |
| 172 |
*gfx_ip_version_major = (ip_info.ip_discovery_version >> 16) & 0xff; |
| 173 |
} else { |
| 174 |
*gfx_ip_version_major = ip_info.hw_ip_version_major; |
| 175 |
} |
| 176 |
|
| 177 |
return r; |
| 178 |
} |
| 179 |
|
| 180 |
static int _amdgpu_device_initialize(int fd, |
| 181 |
uint32_t *major_version, |
| 182 |
uint32_t *minor_version, |
| 183 |
amdgpu_device_handle *device_handle, |
| 184 |
bool deduplicate_device) |
| 185 |
{ |
| 186 |
struct amdgpu_device *dev = NULL; |
| 187 |
uint8_t gfx_ip_version_major = 0; |
| 188 |
drmVersionPtr version; |
| 189 |
int r; |
| 190 |
int flag_auth = 0; |
| 191 |
int flag_authexist=0; |
| 192 |
uint32_t accel_working = 0; |
| 193 |
uint32_t va_mgr_flags = 0; |
| 194 |
|
| 195 |
*device_handle = NULL; |
| 196 |
|
| 197 |
pthread_mutex_lock(&dev_mutex); |
| 198 |
|
| 199 |
r = amdgpu_get_auth(fd, &flag_auth); |
| 200 |
if (r) { |
| 201 |
fprintf(stderr, "%s: amdgpu_get_auth (1) failed (%i)\n", |
| 202 |
__func__, r); |
| 203 |
pthread_mutex_unlock(&dev_mutex); |
| 204 |
return r; |
| 205 |
} |
| 206 |
|
| 207 |
if (deduplicate_device) |
| 208 |
for (dev = dev_list; dev; dev = dev->next) |
| 209 |
if (fd_compare(dev->fd, fd) == 0) |
| 210 |
break; |
| 211 |
|
| 212 |
if (dev) { |
| 213 |
r = amdgpu_get_auth(dev->fd, &flag_authexist); |
| 214 |
if (r) { |
| 215 |
fprintf(stderr, "%s: amdgpu_get_auth (2) failed (%i)\n", |
| 216 |
__func__, r); |
| 217 |
pthread_mutex_unlock(&dev_mutex); |
| 218 |
return r; |
| 219 |
} |
| 220 |
if ((flag_auth) && (!flag_authexist)) { |
| 221 |
dev->flink_fd = fcntl(fd, F_DUPFD_CLOEXEC, 0); |
| 222 |
} |
| 223 |
*major_version = dev->major_version; |
| 224 |
*minor_version = dev->minor_version; |
| 225 |
amdgpu_device_reference(device_handle, dev); |
| 226 |
pthread_mutex_unlock(&dev_mutex); |
| 227 |
return 0; |
| 228 |
} |
| 229 |
|
| 230 |
dev = calloc(1, sizeof(struct amdgpu_device)); |
| 231 |
if (!dev) { |
| 232 |
fprintf(stderr, "%s: calloc failed\n", __func__); |
| 233 |
pthread_mutex_unlock(&dev_mutex); |
| 234 |
return -ENOMEM; |
| 235 |
} |
| 236 |
|
| 237 |
dev->fd = -1; |
| 238 |
dev->flink_fd = -1; |
| 239 |
|
| 240 |
atomic_set(&dev->refcount, 1); |
| 241 |
|
| 242 |
version = drmGetVersion(fd); |
| 243 |
if (version->version_major != 3) { |
| 244 |
fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is " |
| 245 |
"only compatible with 3.x.x.\n", |
| 246 |
__func__, |
| 247 |
version->version_major, |
| 248 |
version->version_minor, |
| 249 |
version->version_patchlevel); |
| 250 |
drmFreeVersion(version); |
| 251 |
r = -EBADF; |
| 252 |
goto cleanup; |
| 253 |
} |
| 254 |
|
| 255 |
dev->fd = fcntl(fd, F_DUPFD_CLOEXEC, 0); |
| 256 |
dev->flink_fd = dev->fd; |
| 257 |
dev->major_version = version->version_major; |
| 258 |
dev->minor_version = version->version_minor; |
| 259 |
drmFreeVersion(version); |
| 260 |
|
| 261 |
pthread_mutex_init(&dev->bo_table_mutex, NULL); |
| 262 |
|
| 263 |
/* Check if acceleration is working. */ |
| 264 |
r = amdgpu_query_info(dev, AMDGPU_INFO_ACCEL_WORKING, 4, &accel_working); |
| 265 |
if (r) { |
| 266 |
fprintf(stderr, "%s: amdgpu_query_info(ACCEL_WORKING) failed (%i)\n", |
| 267 |
__func__, r); |
| 268 |
goto cleanup; |
| 269 |
} |
| 270 |
if (!accel_working) { |
| 271 |
fprintf(stderr, "%s: AMDGPU_INFO_ACCEL_WORKING = 0\n", __func__); |
| 272 |
r = -EBADF; |
| 273 |
goto cleanup; |
| 274 |
} |
| 275 |
|
| 276 |
r = amdgpu_query_gpu_info_init(dev); |
| 277 |
if (r) { |
| 278 |
fprintf(stderr, "%s: amdgpu_query_gpu_info_init failed\n", __func__); |
| 279 |
goto cleanup; |
| 280 |
} |
| 281 |
|
| 282 |
r = amdgpu_query_gfx_level_major(dev, &gfx_ip_version_major); |
| 283 |
if (!r) { |
| 284 |
/* Split the HIGH addr space for GFX6-GFX12, except GFX9 to |
| 285 |
* implement a workaround for SMEM loads with NULL PRT pages. |
| 286 |
* This is silently ignored if querying the GFX level failed. |
| 287 |
*/ |
| 288 |
if (gfx_ip_version_major >= 6 && gfx_ip_version_major <= 12 && |
| 289 |
gfx_ip_version_major != 9) { |
| 290 |
va_mgr_flags |= AMDGPU_VA_MGR_RESERVE_HALF_VA_FOR_PRT; |
| 291 |
} |
| 292 |
} |
| 293 |
|
| 294 |
amdgpu_va_manager_init2(&dev->va_mgr, |
| 295 |
dev->dev_info.virtual_address_offset, |
| 296 |
dev->dev_info.virtual_address_max, |
| 297 |
dev->dev_info.high_va_offset, |
| 298 |
dev->dev_info.high_va_max, |
| 299 |
dev->dev_info.virtual_address_alignment, |
| 300 |
va_mgr_flags); |
| 301 |
|
| 302 |
amdgpu_parse_asic_ids(dev); |
| 303 |
|
| 304 |
*major_version = dev->major_version; |
| 305 |
*minor_version = dev->minor_version; |
| 306 |
*device_handle = dev; |
| 307 |
if (deduplicate_device) { |
| 308 |
dev->next = dev_list; |
| 309 |
dev_list = dev; |
| 310 |
} |
| 311 |
pthread_mutex_unlock(&dev_mutex); |
| 312 |
|
| 313 |
return 0; |
| 314 |
|
| 315 |
cleanup: |
| 316 |
if (dev->fd >= 0) |
| 317 |
close(dev->fd); |
| 318 |
free(dev); |
| 319 |
pthread_mutex_unlock(&dev_mutex); |
| 320 |
return r; |
| 321 |
} |
| 322 |
|
| 323 |
drm_public int amdgpu_device_initialize(int fd, |
| 324 |
uint32_t *major_version, |
| 325 |
uint32_t *minor_version, |
| 326 |
amdgpu_device_handle *device_handle) |
| 327 |
{ |
| 328 |
return _amdgpu_device_initialize(fd, major_version, minor_version, device_handle, true); |
| 329 |
} |
| 330 |
|
| 331 |
drm_public int amdgpu_device_initialize2(int fd, bool deduplicate_device, |
| 332 |
uint32_t *major_version, |
| 333 |
uint32_t *minor_version, |
| 334 |
amdgpu_device_handle *device_handle) |
| 335 |
{ |
| 336 |
return _amdgpu_device_initialize(fd, major_version, minor_version, device_handle, deduplicate_device); |
| 337 |
} |
| 338 |
|
| 339 |
drm_public int amdgpu_device_deinitialize(amdgpu_device_handle dev) |
| 340 |
{ |
| 341 |
pthread_mutex_lock(&dev_mutex); |
| 342 |
amdgpu_device_reference(&dev, NULL); |
| 343 |
pthread_mutex_unlock(&dev_mutex); |
| 344 |
return 0; |
| 345 |
} |
| 346 |
|
| 347 |
drm_public int amdgpu_device_get_fd(amdgpu_device_handle device_handle) |
| 348 |
{ |
| 349 |
return device_handle->fd; |
| 350 |
} |
| 351 |
|
| 352 |
drm_public const char *amdgpu_get_marketing_name(amdgpu_device_handle dev) |
| 353 |
{ |
| 354 |
if (dev->marketing_name) |
| 355 |
return dev->marketing_name; |
| 356 |
else |
| 357 |
return "AMD Radeon Graphics"; |
| 358 |
} |
| 359 |
|
| 360 |
drm_public int amdgpu_query_sw_info(amdgpu_device_handle dev, |
| 361 |
enum amdgpu_sw_info info, |
| 362 |
void *value) |
| 363 |
{ |
| 364 |
uint32_t *val32 = (uint32_t*)value; |
| 365 |
|
| 366 |
switch (info) { |
| 367 |
case amdgpu_sw_info_address32_hi: |
| 368 |
if (dev->va_mgr.vamgr_high_32.va_max) |
| 369 |
*val32 = (dev->va_mgr.vamgr_high_32.va_max - 1) >> 32; |
| 370 |
else |
| 371 |
*val32 = (dev->va_mgr.vamgr_32.va_max - 1) >> 32; |
| 372 |
return 0; |
| 373 |
case amdgpu_sw_info_address_prt_wa_control_bit: |
| 374 |
*val32 = dev->va_mgr.address_prt_wa_control_bit; |
| 375 |
return 0; |
| 376 |
} |
| 377 |
return -EINVAL; |
| 378 |
} |