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_psp.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 struct amdgpu_hive_info {
  35         uint64_t                hive_id;
  36         struct list_head        device_list;
  37 };
  38
  39 static struct amdgpu_hive_info xgmi_hives[AMDGPU_MAX_XGMI_HIVE];
  40 static unsigned hive_count = 0;
  41
  42 static struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev)
  43 {
  44         int i;
  45         struct amdgpu_hive_info *tmp;
  46
  47         if (!adev->gmc.xgmi.hive_id)
  48                 return NULL;
  49         for (i = 0 ; i < hive_count; ++i) {
  50                 tmp = &xgmi_hives[i];
  51                 if (tmp->hive_id == adev->gmc.xgmi.hive_id)
  52                         return tmp;
  53         }
  54         if (i >= AMDGPU_MAX_XGMI_HIVE)
  55                 return NULL;
  56
  57         /* initialize new hive if not exist */
  58         tmp = &xgmi_hives[hive_count++];
  59         tmp->hive_id = adev->gmc.xgmi.hive_id;
  60         INIT_LIST_HEAD(&tmp->device_list);
  61         return tmp;
  62 }
  63
  64 int amdgpu_xgmi_add_device(struct amdgpu_device *adev)
  65 {
  66         struct psp_xgmi_topology_info tmp_topology[AMDGPU_MAX_XGMI_DEVICE_PER_HIVE];
  67         struct amdgpu_hive_info *hive;
  68         struct amdgpu_xgmi      *entry;
  69         struct amdgpu_device    *tmp_adev;
  70
  71         int count = 0, ret = -EINVAL;
  72
  73         if ((adev->asic_type < CHIP_VEGA20) ||
  74                 (adev->flags & AMD_IS_APU) )
  75                 return 0;
  76         adev->gmc.xgmi.device_id = psp_xgmi_get_device_id(&adev->psp);
  77         adev->gmc.xgmi.hive_id = psp_xgmi_get_hive_id(&adev->psp);
  78
  79         memset(&tmp_topology[0], 0, sizeof(tmp_topology));
  80         mutex_lock(&xgmi_mutex);
  81         hive = amdgpu_get_xgmi_hive(adev);
  82         if (!hive)
  83                 goto exit;
  84
  85         list_add_tail(&adev->gmc.xgmi.head, &hive->device_list);
  86         list_for_each_entry(entry, &hive->device_list, head)
  87                 tmp_topology[count++].device_id = entry->device_id;
  88
  89         ret = psp_xgmi_get_topology_info(&adev->psp, count, tmp_topology);
  90         if (ret) {
  91                 dev_err(adev->dev,
  92                         "XGMI: Get topology failure on device %llx, hive %llx, ret %d",
  93                         adev->gmc.xgmi.device_id,
  94                         adev->gmc.xgmi.hive_id, ret);
  95                 goto exit;
  96         }
  97         /* Each psp need to set the latest topology */
  98         list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
  99                 ret = psp_xgmi_set_topology_info(&tmp_adev->psp, count, tmp_topology);
 100                 if (ret) {
 101                         dev_err(tmp_adev->dev,
 102                                 "XGMI: Set topology failure on device %llx, hive %llx, ret %d",
 103                                 tmp_adev->gmc.xgmi.device_id,
 104                                 tmp_adev->gmc.xgmi.hive_id, ret);
 105                         /* To do : continue with some  node failed or disable the  whole  hive */
 106                         break;
 107                 }
 108         }
 109         if (!ret)
 110                 dev_info(adev->dev, "XGMI: Add node %d to hive 0x%llx.\n",
 111                         adev->gmc.xgmi.physical_node_id,
 112                         adev->gmc.xgmi.hive_id);
 113
 114 exit:
 115         mutex_unlock(&xgmi_mutex);
 116         return ret;
 117 }
 118
 119