summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c')
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c100
1 files changed, 69 insertions, 31 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
index 897afbb348c1..0b263a9857c6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
@@ -23,7 +23,7 @@
*/
#include <linux/list.h>
#include "amdgpu.h"
-#include "amdgpu_psp.h"
+#include "amdgpu_xgmi.h"
static DEFINE_MUTEX(xgmi_mutex);
@@ -31,15 +31,16 @@ static DEFINE_MUTEX(xgmi_mutex);
#define AMDGPU_MAX_XGMI_HIVE 8
#define AMDGPU_MAX_XGMI_DEVICE_PER_HIVE 4
-struct amdgpu_hive_info {
- uint64_t hive_id;
- struct list_head device_list;
-};
-
static struct amdgpu_hive_info xgmi_hives[AMDGPU_MAX_XGMI_HIVE];
static unsigned hive_count = 0;
-static struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev)
+
+void *amdgpu_xgmi_hive_try_lock(struct amdgpu_hive_info *hive)
+{
+ return &hive->device_list;
+}
+
+struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev)
{
int i;
struct amdgpu_hive_info *tmp;
@@ -58,62 +59,99 @@ static struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev)
tmp = &xgmi_hives[hive_count++];
tmp->hive_id = adev->gmc.xgmi.hive_id;
INIT_LIST_HEAD(&tmp->device_list);
+ mutex_init(&tmp->hive_lock);
+
return tmp;
}
+int amdgpu_xgmi_update_topology(struct amdgpu_hive_info *hive, struct amdgpu_device *adev)
+{
+ int ret = -EINVAL;
+
+ /* Each psp need to set the latest topology */
+ ret = psp_xgmi_set_topology_info(&adev->psp,
+ hive->number_devices,
+ &hive->topology_info);
+ if (ret)
+ dev_err(adev->dev,
+ "XGMI: Set topology failure on device %llx, hive %llx, ret %d",
+ adev->gmc.xgmi.node_id,
+ adev->gmc.xgmi.hive_id, ret);
+ else
+ dev_info(adev->dev, "XGMI: Set topology for node %d, hive 0x%llx.\n",
+ adev->gmc.xgmi.physical_node_id,
+ adev->gmc.xgmi.hive_id);
+
+ return ret;
+}
+
int amdgpu_xgmi_add_device(struct amdgpu_device *adev)
{
- struct psp_xgmi_topology_info tmp_topology[AMDGPU_MAX_XGMI_DEVICE_PER_HIVE];
+ struct psp_xgmi_topology_info *hive_topology;
struct amdgpu_hive_info *hive;
struct amdgpu_xgmi *entry;
- struct amdgpu_device *tmp_adev;
+ struct amdgpu_device *tmp_adev = NULL;
int count = 0, ret = -EINVAL;
- if ((adev->asic_type < CHIP_VEGA20) ||
- (adev->flags & AMD_IS_APU) )
+ if (!adev->gmc.xgmi.supported)
return 0;
- adev->gmc.xgmi.device_id = psp_xgmi_get_device_id(&adev->psp);
+
+ adev->gmc.xgmi.node_id = psp_xgmi_get_node_id(&adev->psp);
adev->gmc.xgmi.hive_id = psp_xgmi_get_hive_id(&adev->psp);
- memset(&tmp_topology[0], 0, sizeof(tmp_topology));
mutex_lock(&xgmi_mutex);
hive = amdgpu_get_xgmi_hive(adev);
if (!hive)
goto exit;
+ hive_topology = &hive->topology_info;
+
list_add_tail(&adev->gmc.xgmi.head, &hive->device_list);
list_for_each_entry(entry, &hive->device_list, head)
- tmp_topology[count++].device_id = entry->device_id;
+ hive_topology->nodes[count++].node_id = entry->node_id;
+ hive->number_devices = count;
- ret = psp_xgmi_get_topology_info(&adev->psp, count, tmp_topology);
- if (ret) {
- dev_err(adev->dev,
- "XGMI: Get topology failure on device %llx, hive %llx, ret %d",
- adev->gmc.xgmi.device_id,
- adev->gmc.xgmi.hive_id, ret);
- goto exit;
- }
- /* Each psp need to set the latest topology */
+ /* Each psp need to get the latest topology */
list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
- ret = psp_xgmi_set_topology_info(&tmp_adev->psp, count, tmp_topology);
+ ret = psp_xgmi_get_topology_info(&tmp_adev->psp, count, hive_topology);
if (ret) {
dev_err(tmp_adev->dev,
- "XGMI: Set topology failure on device %llx, hive %llx, ret %d",
- tmp_adev->gmc.xgmi.device_id,
+ "XGMI: Get topology failure on device %llx, hive %llx, ret %d",
+ tmp_adev->gmc.xgmi.node_id,
tmp_adev->gmc.xgmi.hive_id, ret);
- /* To do : continue with some node failed or disable the whole hive */
+ /* To do : continue with some node failed or disable the whole hive */
break;
}
}
- if (!ret)
- dev_info(adev->dev, "XGMI: Add node %d to hive 0x%llx.\n",
- adev->gmc.xgmi.physical_node_id,
- adev->gmc.xgmi.hive_id);
+
+ list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
+ ret = amdgpu_xgmi_update_topology(hive, tmp_adev);
+ if (ret)
+ break;
+ }
exit:
mutex_unlock(&xgmi_mutex);
return ret;
}
+void amdgpu_xgmi_remove_device(struct amdgpu_device *adev)
+{
+ struct amdgpu_hive_info *hive;
+ if (!adev->gmc.xgmi.supported)
+ return;
+
+ mutex_lock(&xgmi_mutex);
+
+ hive = amdgpu_get_xgmi_hive(adev);
+ if (!hive)
+ goto exit;
+
+ if (!(hive->number_devices--))
+ mutex_destroy(&hive->hive_lock);
+
+exit:
+ mutex_unlock(&xgmi_mutex);
+}