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
  38 void *amdgpu_xgmi_hive_try_lock(struct amdgpu_hive_info *hive)
  39 {
  40         return &hive->device_list;
  41 }
  42
  43 struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev, int lock)
  44 {
  45         int i;
  46         struct amdgpu_hive_info *tmp;
  47
  48         if (!adev->gmc.xgmi.hive_id)
  49                 return NULL;
  50
  51         mutex_lock(&xgmi_mutex);
  52
  53         for (i = 0 ; i < hive_count; ++i) {
  54                 tmp = &xgmi_hives[i];
  55                 if (tmp->hive_id == adev->gmc.xgmi.hive_id) {
  56                         if (lock)
  57                                 mutex_lock(&tmp->hive_lock);
  58                         mutex_unlock(&xgmi_mutex);
  59                         return tmp;
  60                 }
  61         }
  62         if (i >= AMDGPU_MAX_XGMI_HIVE) {
  63                 mutex_unlock(&xgmi_mutex);
  64                 return NULL;
  65         }
  66
  67         /* initialize new hive if not exist */
  68         tmp = &xgmi_hives[hive_count++];
  69         tmp->hive_id = adev->gmc.xgmi.hive_id;
  70         INIT_LIST_HEAD(&tmp->device_list);
  71         mutex_init(&tmp->hive_lock);
  72         mutex_init(&tmp->reset_lock);
  73         if (lock)
  74                 mutex_lock(&tmp->hive_lock);
  75
  76         mutex_unlock(&xgmi_mutex);
  77
  78         return tmp;
  79 }
  80
  81 int amdgpu_xgmi_update_topology(struct amdgpu_hive_info *hive, struct amdgpu_device *adev)
  82 {
  83         int ret = -EINVAL;
  84
  85         /* Each psp need to set the latest topology */
  86         ret = psp_xgmi_set_topology_info(&adev->psp,
  87                                          hive->number_devices,
  88                                          &hive->topology_info);
  89         if (ret)
  90                 dev_err(adev->dev,
  91                         "XGMI: Set topology failure on device %llx, hive %llx, ret %d",
  92                         adev->gmc.xgmi.node_id,
  93                         adev->gmc.xgmi.hive_id, ret);
  94
  95         return ret;
  96 }
  97
  98 int amdgpu_xgmi_add_device(struct amdgpu_device *adev)
  99 {
 100         struct psp_xgmi_topology_info *hive_topology;
 101         struct amdgpu_hive_info *hive;
 102         struct amdgpu_xgmi      *entry;
 103         struct amdgpu_device *tmp_adev = NULL;
 104
 105         int count = 0, ret = -EINVAL;
 106
 107         if (!adev->gmc.xgmi.supported)
 108                 return 0;
 109
 110         ret = psp_xgmi_get_node_id(&adev->psp, &adev->gmc.xgmi.node_id);
 111         if (ret) {
 112                 dev_err(adev->dev,
 113                         "XGMI: Failed to get node id\n");
 114                 return ret;
 115         }
 116
 117         ret = psp_xgmi_get_hive_id(&adev->psp, &adev->gmc.xgmi.hive_id);
 118         if (ret) {
 119                 dev_err(adev->dev,
 120                         "XGMI: Failed to get hive id\n");
 121                 return ret;
 122         }
 123
 124         hive = amdgpu_get_xgmi_hive(adev, 1);
 125         if (!hive) {
 126                 ret = -EINVAL;
 127                 dev_err(adev->dev,
 128                         "XGMI: node 0x%llx, can not match hive 0x%llx in the hive list.\n",
 129                         adev->gmc.xgmi.node_id, adev->gmc.xgmi.hive_id);
 130                 goto exit;
 131         }
 132
 133         hive_topology = &hive->topology_info;
 134
 135         list_add_tail(&adev->gmc.xgmi.head, &hive->device_list);
 136         list_for_each_entry(entry, &hive->device_list, head)
 137                 hive_topology->nodes[count++].node_id = entry->node_id;
 138         hive->number_devices = count;
 139
 140         /* Each psp need to get the latest topology */
 141         list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
 142                 ret = psp_xgmi_get_topology_info(&tmp_adev->psp, count, hive_topology);
 143                 if (ret) {
 144                         dev_err(tmp_adev->dev,
 145                                 "XGMI: Get topology failure on device %llx, hive %llx, ret %d",
 146                                 tmp_adev->gmc.xgmi.node_id,
 147                                 tmp_adev->gmc.xgmi.hive_id, ret);
 148                         /* To do : continue with some node failed or disable the whole hive */
 149                         break;
 150                 }
 151         }
 152
 153         list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
 154                 ret = amdgpu_xgmi_update_topology(hive, tmp_adev);
 155                 if (ret)
 156                         break;
 157         }
 158
 159         dev_info(adev->dev, "XGMI: Add node %d, hive 0x%llx.\n",
 160                  adev->gmc.xgmi.physical_node_id, adev->gmc.xgmi.hive_id);
 161
 162         mutex_unlock(&hive->hive_lock);
 163 exit:
 164         return ret;
 165 }
 166
 167 void amdgpu_xgmi_remove_device(struct amdgpu_device *adev)
 168 {
 169         struct amdgpu_hive_info *hive;
 170
 171         if (!adev->gmc.xgmi.supported)
 172                 return;
 173
 174         hive = amdgpu_get_xgmi_hive(adev, 1);
 175         if (!hive)
 176                 return;
 177
 178         if (!(hive->number_devices--)) {
 179                 mutex_destroy(&hive->hive_lock);
 180                 mutex_destroy(&hive->reset_lock);
 181         } else {
 182                 mutex_unlock(&hive->hive_lock);
 183         }
 184 }