1 files changed, 75 insertions, 44 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
index e8b8f28c2f72..91817a31f3e1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
@@ -32,6 +32,8 @@
 #include "wafl/wafl2_4_0_0_smn.h"
 #include "wafl/wafl2_4_0_0_sh_mask.h"
 
+#include "amdgpu_reset.h"
+
 #define smnPCS_XGMI23_PCS_ERROR_STATUS   0x11a01210
 #define smnPCS_XGMI3X16_PCS_ERROR_STATUS 0x11a0020c
 #define smnPCS_GOPX1_PCS_ERROR_STATUS    0x12200210
@@ -227,6 +229,9 @@ static void amdgpu_xgmi_hive_release(struct kobject *kobj)
 	struct amdgpu_hive_info *hive = container_of(
 		kobj, struct amdgpu_hive_info, kobj);
 
+	amdgpu_reset_put_reset_domain(hive->reset_domain);
+	hive->reset_domain = NULL;
+
 	mutex_destroy(&hive->hive_lock);
 	kfree(hive);
 }
@@ -398,15 +403,35 @@ struct amdgpu_hive_info *amdgpu_get_xgmi_hive(struct amdgpu_device *adev)
 		goto pro_end;
 	}
 
+	/**
+	 * Avoid recreating reset domain when hive is reconstructed for the case
+	 * of reset the devices in the XGMI hive during probe for SRIOV
+	 * See https://www.spinics.net/lists/amd-gfx/msg58836.html
+	 */
+	if (adev->reset_domain->type != XGMI_HIVE) {
+		hive->reset_domain = amdgpu_reset_create_reset_domain(XGMI_HIVE, "amdgpu-reset-hive");
+			if (!hive->reset_domain) {
+				dev_err(adev->dev, "XGMI: failed initializing reset domain for xgmi hive\n");
+				ret = -ENOMEM;
+				kobject_put(&hive->kobj);
+				kfree(hive);
+				hive = NULL;
+				goto pro_end;
+			}
+	} else {
+		amdgpu_reset_get_reset_domain(adev->reset_domain);
+		hive->reset_domain = adev->reset_domain;
+	}
+
 	hive->hive_id = adev->gmc.xgmi.hive_id;
 	INIT_LIST_HEAD(&hive->device_list);
 	INIT_LIST_HEAD(&hive->node);
 	mutex_init(&hive->hive_lock);
-	atomic_set(&hive->in_reset, 0);
 	atomic_set(&hive->number_devices, 0);
 	task_barrier_init(&hive->tb);
 	hive->pstate = AMDGPU_XGMI_PSTATE_UNKNOWN;
 	hive->hi_req_gpu = NULL;
+
 	/*
 	 * hive pstate on boot is high in vega20 so we have to go to low
 	 * pstate on after boot.
@@ -732,53 +757,22 @@ int amdgpu_xgmi_remove_device(struct amdgpu_device *adev)
 	return psp_xgmi_terminate(&adev->psp);
 }
 
-static int amdgpu_xgmi_ras_late_init(struct amdgpu_device *adev)
+static int amdgpu_xgmi_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
 {
-	int r;
-	struct ras_ih_if ih_info = {
-		.cb = NULL,
-	};
-	struct ras_fs_if fs_info = {
-		.sysfs_name = "xgmi_wafl_err_count",
-	};
-
 	if (!adev->gmc.xgmi.supported ||
 	    adev->gmc.xgmi.num_physical_nodes == 0)
 		return 0;
 
-	adev->gmc.xgmi.ras_funcs->reset_ras_error_count(adev);
-
-	if (!adev->gmc.xgmi.ras_if) {
-		adev->gmc.xgmi.ras_if = kmalloc(sizeof(struct ras_common_if), GFP_KERNEL);
-		if (!adev->gmc.xgmi.ras_if)
-			return -ENOMEM;
-		adev->gmc.xgmi.ras_if->block = AMDGPU_RAS_BLOCK__XGMI_WAFL;
-		adev->gmc.xgmi.ras_if->type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
-		adev->gmc.xgmi.ras_if->sub_block_index = 0;
-	}
-	ih_info.head = fs_info.head = *adev->gmc.xgmi.ras_if;
-	r = amdgpu_ras_late_init(adev, adev->gmc.xgmi.ras_if,
-				 &fs_info, &ih_info);
-	if (r || !amdgpu_ras_is_supported(adev, adev->gmc.xgmi.ras_if->block)) {
-		kfree(adev->gmc.xgmi.ras_if);
-		adev->gmc.xgmi.ras_if = NULL;
-	}
+	adev->gmc.xgmi.ras->ras_block.hw_ops->reset_ras_error_count(adev);
 
-	return r;
+	return amdgpu_ras_block_late_init(adev, ras_block);
 }
 
 static void amdgpu_xgmi_ras_fini(struct amdgpu_device *adev)
 {
 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__XGMI_WAFL) &&
-			adev->gmc.xgmi.ras_if) {
-		struct ras_common_if *ras_if = adev->gmc.xgmi.ras_if;
-		struct ras_ih_if ih_info = {
-			.cb = NULL,
-		};
-
-		amdgpu_ras_late_fini(adev, ras_if, &ih_info);
-		kfree(ras_if);
-	}
+			adev->gmc.xgmi.ras_if)
+		amdgpu_ras_block_late_fini(adev, adev->gmc.xgmi.ras_if);
 }
 
 uint64_t amdgpu_xgmi_get_relative_phy_addr(struct amdgpu_device *adev,
@@ -865,7 +859,7 @@ static int amdgpu_xgmi_query_pcs_error_status(struct amdgpu_device *adev,
 	return 0;
 }
 
-static int amdgpu_xgmi_query_ras_error_count(struct amdgpu_device *adev,
+static void amdgpu_xgmi_query_ras_error_count(struct amdgpu_device *adev,
 					     void *ras_error_status)
 {
 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
@@ -874,7 +868,7 @@ static int amdgpu_xgmi_query_ras_error_count(struct amdgpu_device *adev,
 	uint32_t ue_cnt = 0, ce_cnt = 0;
 
 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__XGMI_WAFL))
-		return -EINVAL;
+		return ;
 
 	err_data->ue_count = 0;
 	err_data->ce_count = 0;
@@ -940,17 +934,54 @@ static int amdgpu_xgmi_query_ras_error_count(struct amdgpu_device *adev,
 		break;
 	}
 
-	adev->gmc.xgmi.ras_funcs->reset_ras_error_count(adev);
+	adev->gmc.xgmi.ras->ras_block.hw_ops->reset_ras_error_count(adev);
 
 	err_data->ue_count += ue_cnt;
 	err_data->ce_count += ce_cnt;
+}
 
-	return 0;
+/* Trigger XGMI/WAFL error */
+static int amdgpu_ras_error_inject_xgmi(struct amdgpu_device *adev,  void *inject_if)
+{
+	int ret = 0;
+	struct ta_ras_trigger_error_input *block_info =
+				(struct ta_ras_trigger_error_input *)inject_if;
+
+	if (amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_DISALLOW))
+		dev_warn(adev->dev, "Failed to disallow df cstate");
+
+	if (amdgpu_dpm_allow_xgmi_power_down(adev, false))
+		dev_warn(adev->dev, "Failed to disallow XGMI power down");
+
+	ret = psp_ras_trigger_error(&adev->psp, block_info);
+
+	if (amdgpu_ras_intr_triggered())
+		return ret;
+
+	if (amdgpu_dpm_allow_xgmi_power_down(adev, true))
+		dev_warn(adev->dev, "Failed to allow XGMI power down");
+
+	if (amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_ALLOW))
+		dev_warn(adev->dev, "Failed to allow df cstate");
+
+	return ret;
 }
 
-const struct amdgpu_xgmi_ras_funcs xgmi_ras_funcs = {
-	.ras_late_init = amdgpu_xgmi_ras_late_init,
-	.ras_fini = amdgpu_xgmi_ras_fini,
+struct amdgpu_ras_block_hw_ops  xgmi_ras_hw_ops = {
 	.query_ras_error_count = amdgpu_xgmi_query_ras_error_count,
 	.reset_ras_error_count = amdgpu_xgmi_reset_ras_error_count,
+	.ras_error_inject = amdgpu_ras_error_inject_xgmi,
+};
+
+struct amdgpu_xgmi_ras xgmi_ras = {
+	.ras_block = {
+		.ras_comm = {
+			.name = "xgmi_wafl",
+			.block = AMDGPU_RAS_BLOCK__XGMI_WAFL,
+			.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
+		},
+		.hw_ops = &xgmi_ras_hw_ops,
+		.ras_late_init = amdgpu_xgmi_ras_late_init,
+		.ras_fini = amdgpu_xgmi_ras_fini,
+	},
 };