drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c

   1 /*
   2  * Copyright 2018 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 #include <linux/list.h>
  25 #include "amdgpu.h"
  26 #include "amdgpu_xgmi.h"
  27
  28
  29 static DEFINE_MUTEX(xgmi_mutex);
  30
  31 #define AMDGPU_MAX_XGMI_HIVE                    8
  32 #define AMDGPU_MAX_XGMI_DEVICE_PER_HIVE         4
  33
  34 static struct amdgpu_hive_info xgmi_hives[AMDGPU_MAX_XGMI_HIVE];
  35 static unsigned hive_count = 0;
  36
  37 void *amdgpu_xgmi_hive_try_lock(struct amdgpu_hive_info *hive)
  38 {
  39         return &hive->device_list;
  40 }
  41
  42 static ssize_t amdgpu_xgmi_show_hive_id(struct device *dev,
  43                 struct device_attribute *attr, char *buf)
  44 {
  45         struct amdgpu_hive_info *hive =
  46                         container_of(attr, struct amdgpu_hive_info, dev_attr);
  47
  48         return snprintf(buf, PAGE_SIZE, "%llu\n", hive->hive_id);
  49 }
  50
  51 static int amdgpu_xgmi_sysfs_create(struct amdgpu_device *adev,
  52                                     struct amdgpu_hive_info *hive)
  53 {
  54         int ret = 0;
  55
  56         if (WARN_ON(hive->kobj))
  57                 return -EINVAL;
  58
  59         hive->kobj = kobject_create_and_add("xgmi_hive_info", &adev->dev->kobj);
  60         if (!hive->kobj) {
  61                 dev_err(adev->dev, "XGMI: Failed to allocate sysfs entry!\n");
  62                 return -EINVAL;
  63         }
  64
  65         hive->dev_attr = (struct device_attribute) {
  66                 .attr = {
  67                         .name = "xgmi_hive_id",
  68                         .mode = S_IRUGO,
  69
  70                 },
  71                 .show = amdgpu_xgmi_show_hive_id,
  72         };
  73
  74         ret = sysfs_create_file(hive->kobj, &hive->dev_attr.attr);
  75         if (ret) {
  76                 dev_err(adev->dev, "XGMI: Failed to create device file xgmi_hive_id\n");
  77                 kobject_del(hive->kobj);
  78                 kobject_put(hive->kobj);
  79                 hive->kobj = NULL;
  80         }
  81
  82         return ret;
  83 }
  84
  85 static void amdgpu_xgmi_sysfs_destroy(struct amdgpu_device *adev,
  86                                     struct amdgpu_hive_info *hive)
  87 {
  88         sysfs_remove_file(hive->kobj, &hive->dev_attr.attr);
  89         kobject_del(hive->kobj);
  90         kobject_put(hive->kobj);
  91         hive->kobj = NULL;
  92 }
  93
  94 static ssize_t amdgpu_xgmi_show_device_id(struct device *dev,
  95                                      struct device_attribute *attr,
  96                                      char *buf)
  97 {
  98         struct drm_device *ddev = dev_get_drvdata(dev);
  99         struct amdgpu_device *adev = ddev->dev_private;
 100
 101         return snprintf(buf, PAGE_SIZE, "%llu\n", adev->gmc.xgmi.node_id);
 102
 103 }
 104
 105
 106 static DEVICE_ATTR(xgmi_device_id, S_IRUGO, amdgpu_xgmi_show_device_id, NULL);
 107
 108
 109 static int amdgpu_xgmi_sysfs_add_dev_info(struct amdgpu_device *adev,
 110                                          struct amdgpu_hive_info *hive)
 111 {
 112         int ret = 0;
 113         char node[10] = { 0 };
 114
 115         /* Create xgmi device id file */
 116         ret = device_create_file(adev->dev, &dev_attr_xgmi_device_id);
 117         if (ret) {
 118                 dev_err(adev->dev, "XGMI: Failed to create device file xgmi_device_id\n");
 119                 return ret;
 120         }
 121
 122         /* Create sysfs link to hive info folder on the first device */
 123         if (adev != hive->adev) {
 124                 ret = sysfs_create_link(&adev->dev->kobj, hive->kobj,
 125                                         "xgmi_hive_info");
 126                 if (ret) {
 127                         dev_err(adev->dev, "XGMI: Failed to create link to hive info");
 128                         goto remove_file;
 129                 }
 130         }
 131
 132         sprintf(node, "node%d", hive->number_devices);
 133         /* Create sysfs link form the hive folder to yourself */
 134         ret = sysfs_create_link(hive->kobj, &adev->dev->kobj, node);
 135         if (ret) {
 136                 dev_err(adev->dev, "XGMI: Failed to create link from hive info");
 137                 goto remove_link;
 138         }
 139
 140         goto success;
 141
 142
 143 remove_link:
 144         sysfs_remove_link(&adev->dev->kobj, adev->ddev->unique);
 145
 146 remove_file:
 147         device_remove_file(adev->dev, &dev_attr_xgmi_device_id);
 148
 149 success:
 150         return ret;
 151 }
 152
 153 static void amdgpu_xgmi_sysfs_rem_dev_info(struct amdgpu_device *adev,
 154                                           struct amdgpu_hive_info *hive)
 155 {
 156         device_remove_file(adev->dev, &dev_attr_xgmi_device_id);
 157         sysfs_remove_link(&adev->dev->kobj, adev->ddev->unique);
 158         sysfs_remove_link(hive->kobj, adev->ddev->unique);
 159 }
 160
 161
 162
 163 struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev, int lock)
 164 {
 165         int i;
 166         struct amdgpu_hive_info *tmp;
 167
 168         if (!adev->gmc.xgmi.hive_id)
 169                 return NULL;
 170
 171         mutex_lock(&xgmi_mutex);
 172
 173         for (i = 0 ; i < hive_count; ++i) {
 174                 tmp = &xgmi_hives[i];
 175                 if (tmp->hive_id == adev->gmc.xgmi.hive_id) {
 176                         if (lock)
 177                                 mutex_lock(&tmp->hive_lock);
 178                         mutex_unlock(&xgmi_mutex);
 179                         return tmp;
 180                 }
 181         }
 182         if (i >= AMDGPU_MAX_XGMI_HIVE) {
 183                 mutex_unlock(&xgmi_mutex);
 184                 return NULL;
 185         }
 186
 187         /* initialize new hive if not exist */
 188         tmp = &xgmi_hives[hive_count++];
 189
 190         if (amdgpu_xgmi_sysfs_create(adev, tmp)) {
 191                 mutex_unlock(&xgmi_mutex);
 192                 return NULL;
 193         }
 194
 195         tmp->adev = adev;
 196         tmp->hive_id = adev->gmc.xgmi.hive_id;
 197         INIT_LIST_HEAD(&tmp->device_list);
 198         mutex_init(&tmp->hive_lock);
 199         mutex_init(&tmp->reset_lock);
 200
 201         if (lock)
 202                 mutex_lock(&tmp->hive_lock);
 203
 204         mutex_unlock(&xgmi_mutex);
 205
 206         return tmp;
 207 }
 208
 209 int amdgpu_xgmi_update_topology(struct amdgpu_hive_info *hive, struct amdgpu_device *adev)
 210 {
 211         int ret = -EINVAL;
 212
 213         /* Each psp need to set the latest topology */
 214         ret = psp_xgmi_set_topology_info(&adev->psp,
 215                                          hive->number_devices,
 216                                          &hive->topology_info);
 217         if (ret)
 218                 dev_err(adev->dev,
 219                         "XGMI: Set topology failure on device %llx, hive %llx, ret %d",
 220                         adev->gmc.xgmi.node_id,
 221                         adev->gmc.xgmi.hive_id, ret);
 222
 223         return ret;
 224 }
 225
 226 int amdgpu_xgmi_add_device(struct amdgpu_device *adev)
 227 {
 228         struct psp_xgmi_topology_info *hive_topology;
 229         struct amdgpu_hive_info *hive;
 230         struct amdgpu_xgmi      *entry;
 231         struct amdgpu_device *tmp_adev = NULL;
 232
 233         int count = 0, ret = -EINVAL;
 234
 235         if (!adev->gmc.xgmi.supported)
 236                 return 0;
 237
 238         ret = psp_xgmi_get_node_id(&adev->psp, &adev->gmc.xgmi.node_id);
 239         if (ret) {
 240                 dev_err(adev->dev,
 241                         "XGMI: Failed to get node id\n");
 242                 return ret;
 243         }
 244
 245         ret = psp_xgmi_get_hive_id(&adev->psp, &adev->gmc.xgmi.hive_id);
 246         if (ret) {
 247                 dev_err(adev->dev,
 248                         "XGMI: Failed to get hive id\n");
 249                 return ret;
 250         }
 251
 252         hive = amdgpu_get_xgmi_hive(adev, 1);
 253         if (!hive) {
 254                 ret = -EINVAL;
 255                 dev_err(adev->dev,
 256                         "XGMI: node 0x%llx, can not match hive 0x%llx in the hive list.\n",
 257                         adev->gmc.xgmi.node_id, adev->gmc.xgmi.hive_id);
 258                 goto exit;
 259         }
 260
 261         hive_topology = &hive->topology_info;
 262
 263         list_add_tail(&adev->gmc.xgmi.head, &hive->device_list);
 264         list_for_each_entry(entry, &hive->device_list, head)
 265                 hive_topology->nodes[count++].node_id = entry->node_id;
 266         hive->number_devices = count;
 267
 268         /* Each psp need to get the latest topology */
 269         list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
 270                 ret = psp_xgmi_get_topology_info(&tmp_adev->psp, count, hive_topology);
 271                 if (ret) {
 272                         dev_err(tmp_adev->dev,
 273                                 "XGMI: Get topology failure on device %llx, hive %llx, ret %d",
 274                                 tmp_adev->gmc.xgmi.node_id,
 275                                 tmp_adev->gmc.xgmi.hive_id, ret);
 276                         /* To do : continue with some node failed or disable the whole hive */
 277                         break;
 278                 }
 279         }
 280
 281         list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
 282                 ret = amdgpu_xgmi_update_topology(hive, tmp_adev);
 283                 if (ret)
 284                         break;
 285         }
 286
 287         if (!ret)
 288                 ret = amdgpu_xgmi_sysfs_add_dev_info(adev, hive);
 289
 290         if (!ret)
 291                 dev_info(adev->dev, "XGMI: Add node %d, hive 0x%llx.\n",
 292                          adev->gmc.xgmi.physical_node_id, adev->gmc.xgmi.hive_id);
 293         else
 294                 dev_err(adev->dev, "XGMI: Failed to add node %d, hive 0x%llx ret: %d\n",
 295                         adev->gmc.xgmi.physical_node_id, adev->gmc.xgmi.hive_id,
 296                         ret);
 297
 298
 299         mutex_unlock(&hive->hive_lock);
 300 exit:
 301         return ret;
 302 }
 303
 304 void amdgpu_xgmi_remove_device(struct amdgpu_device *adev)
 305 {
 306         struct amdgpu_hive_info *hive;
 307
 308         if (!adev->gmc.xgmi.supported)
 309                 return;
 310
 311         hive = amdgpu_get_xgmi_hive(adev, 1);
 312         if (!hive)
 313                 return;
 314
 315         if (!(hive->number_devices--)) {
 316                 amdgpu_xgmi_sysfs_destroy(adev, hive);
 317                 mutex_destroy(&hive->hive_lock);
 318                 mutex_destroy(&hive->reset_lock);
 319         } else {
 320                 amdgpu_xgmi_sysfs_rem_dev_info(adev, hive);
 321                 mutex_unlock(&hive->hive_lock);
 322         }
 323 }