drm/amdgpu: Introduce VF critical region check for RAS poison injection

The SRIOV guest send requet to host to check whether the poison
injection address is in VF critical region or not via mabox.

Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Shravan Kumar Gande <Shravankumar.Gande@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Author: Xiang Liu

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

@@ -828,11 +828,14 @@ static void amdgpu_virt_init_ras(struct amdgpu_device *adev)
 {
 	ratelimit_state_init(&adev->virt.ras.ras_error_cnt_rs, 5 * HZ, 1);
 	ratelimit_state_init(&adev->virt.ras.ras_cper_dump_rs, 5 * HZ, 1);
+	ratelimit_state_init(&adev->virt.ras.ras_chk_criti_rs, 5 * HZ, 1);
 
 	ratelimit_set_flags(&adev->virt.ras.ras_error_cnt_rs,
 			    RATELIMIT_MSG_ON_RELEASE);
 	ratelimit_set_flags(&adev->virt.ras.ras_cper_dump_rs,
 			    RATELIMIT_MSG_ON_RELEASE);
+	ratelimit_set_flags(&adev->virt.ras.ras_chk_criti_rs,
+			    RATELIMIT_MSG_ON_RELEASE);
 
 	mutex_init(&adev->virt.ras.ras_telemetry_mutex);
 
@@ -1501,3 +1504,55 @@ void amdgpu_virt_request_bad_pages(struct amdgpu_device *adev)
 	if (virt->ops && virt->ops->req_bad_pages)
 		virt->ops->req_bad_pages(adev);
 }
+
+static int amdgpu_virt_cache_chk_criti_hit(struct amdgpu_device *adev,
+					   struct amdsriov_ras_telemetry *host_telemetry,
+					   bool *hit)
+{
+	struct amd_sriov_ras_chk_criti *tmp = NULL;
+	uint32_t checksum, used_size;
+
+	checksum = host_telemetry->header.checksum;
+	used_size = host_telemetry->header.used_size;
+
+	if (used_size > (AMD_SRIOV_RAS_TELEMETRY_SIZE_KB << 10))
+		return 0;
+
+	tmp = kmemdup(&host_telemetry->body.chk_criti, used_size, GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+
+	if (checksum != amd_sriov_msg_checksum(tmp, used_size, 0, 0))
+		goto out;
+
+	if (hit)
+		*hit = tmp->hit ? true : false;
+
+out:
+	kfree(tmp);
+
+	return 0;
+}
+
+int amdgpu_virt_check_vf_critical_region(struct amdgpu_device *adev, u64 addr, bool *hit)
+{
+	struct amdgpu_virt *virt = &adev->virt;
+	int r = -EPERM;
+
+	if (!virt->ops || !virt->ops->req_ras_chk_criti)
+		return -EOPNOTSUPP;
+
+	/* Host allows 15 ras telemetry requests per 60 seconds. Afterwhich, the Host
+	 * will ignore incoming guest messages. Ratelimit the guest messages to
+	 * prevent guest self DOS.
+	 */
+	if (__ratelimit(&virt->ras.ras_chk_criti_rs)) {
+		mutex_lock(&virt->ras.ras_telemetry_mutex);
+		if (!virt->ops->req_ras_chk_criti(adev, addr))
+			r = amdgpu_virt_cache_chk_criti_hit(
+				adev, virt->fw_reserve.ras_telemetry, hit);
+		mutex_unlock(&virt->ras.ras_telemetry_mutex);
+	}
+
+	return r;
+}

drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h

@@ -98,6 +98,7 @@ struct amdgpu_virt_ops {
 	int (*req_ras_err_count)(struct amdgpu_device *adev);
 	int (*req_ras_cper_dump)(struct amdgpu_device *adev, u64 vf_rptr);
 	int (*req_bad_pages)(struct amdgpu_device *adev);
+	int (*req_ras_chk_criti)(struct amdgpu_device *adev, u64 addr);
 };
 
 /*
@@ -252,6 +253,7 @@ struct amdgpu_virt_ras_err_handler_data {
 struct amdgpu_virt_ras {
 	struct ratelimit_state ras_error_cnt_rs;
 	struct ratelimit_state ras_cper_dump_rs;
+	struct ratelimit_state ras_chk_criti_rs;
 	struct mutex ras_telemetry_mutex;
 	uint64_t cper_rptr;
 };
@@ -453,4 +455,5 @@ int amdgpu_virt_ras_telemetry_post_reset(struct amdgpu_device *adev);
 bool amdgpu_virt_ras_telemetry_block_en(struct amdgpu_device *adev,
 					enum amdgpu_ras_block block);
 void amdgpu_virt_request_bad_pages(struct amdgpu_device *adev);
+int amdgpu_virt_check_vf_critical_region(struct amdgpu_device *adev, u64 addr, bool *hit);
 #endif

drivers/gpu/drm/amd/amdgpu/amdgv_sriovmsg.h

@@ -405,12 +405,17 @@ struct amd_sriov_ras_cper_dump {
 	uint32_t buf[];
 };
 
+struct amd_sriov_ras_chk_criti {
+	uint32_t hit;
+};
+
 struct amdsriov_ras_telemetry {
 	struct amd_sriov_ras_telemetry_header header;
 
 	union {
 		struct amd_sriov_ras_telemetry_error_count error_count;
 		struct amd_sriov_ras_cper_dump cper_dump;
+		struct amd_sriov_ras_chk_criti chk_criti;
 	} body;
 };
 

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

@@ -202,6 +202,9 @@ static int xgpu_nv_send_access_requests_with_param(struct amdgpu_device *adev,
 	case IDH_REQ_RAS_CPER_DUMP:
 		event = IDH_RAS_CPER_DUMP_READY;
 		break;
+	case IDH_REQ_RAS_CHK_CRITI:
+		event = IDH_REQ_RAS_CHK_CRITI_READY;
+		break;
 	default:
 		break;
 	}
@@ -556,6 +559,16 @@ static int xgpu_nv_req_ras_bad_pages(struct amdgpu_device *adev)
 	return xgpu_nv_send_access_requests(adev, IDH_REQ_RAS_BAD_PAGES);
 }
 
+static int xgpu_nv_check_vf_critical_region(struct amdgpu_device *adev, u64 addr)
+{
+	uint32_t addr_hi, addr_lo;
+
+	addr_hi = (uint32_t)(addr >> 32);
+	addr_lo = (uint32_t)(addr & 0xFFFFFFFF);
+	return xgpu_nv_send_access_requests_with_param(
+		adev, IDH_REQ_RAS_CHK_CRITI, addr_hi, addr_lo, 0);
+}
+
 const struct amdgpu_virt_ops xgpu_nv_virt_ops = {
 	.req_full_gpu	= xgpu_nv_request_full_gpu_access,
 	.rel_full_gpu	= xgpu_nv_release_full_gpu_access,
@@ -569,4 +582,5 @@ const struct amdgpu_virt_ops xgpu_nv_virt_ops = {
 	.req_ras_err_count = xgpu_nv_req_ras_err_count,
 	.req_ras_cper_dump = xgpu_nv_req_ras_cper_dump,
 	.req_bad_pages = xgpu_nv_req_ras_bad_pages,
+	.req_ras_chk_criti = xgpu_nv_check_vf_critical_region
 };

drivers/gpu/drm/amd/amdgpu/mxgpu_nv.h

@@ -43,6 +43,7 @@ enum idh_request {
 	IDH_REQ_RAS_ERROR_COUNT = 203,
 	IDH_REQ_RAS_CPER_DUMP = 204,
 	IDH_REQ_RAS_BAD_PAGES = 205,
+	IDH_REQ_RAS_CHK_CRITI = 206
 };
 
 enum idh_event {
@@ -62,6 +63,7 @@ enum idh_event {
 	IDH_RAS_BAD_PAGES_READY = 15,
 	IDH_RAS_BAD_PAGES_NOTIFICATION = 16,
 	IDH_UNRECOV_ERR_NOTIFICATION = 17,
+	IDH_REQ_RAS_CHK_CRITI_READY = 18,
 
 	IDH_TEXT_MESSAGE = 255,
 };