+ "Reset device on internal errors if non-zero (default 1)");
+
+static int read_vendor_id(struct mlx4_dev *dev)
+{
+ u16 vendor_id = 0;
+ int ret;
+
+ ret = pci_read_config_word(dev->persist->pdev, 0, &vendor_id);
+ if (ret) {
+ mlx4_err(dev, "Failed to read vendor ID, ret=%d\n", ret);
+ return ret;
+ }
+
+ if (vendor_id == 0xffff) {
+ mlx4_err(dev, "PCI can't be accessed to read vendor id\n");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int mlx4_reset_master(struct mlx4_dev *dev)
+{
+ int err = 0;
+
+ if (mlx4_is_master(dev))
+ mlx4_report_internal_err_comm_event(dev);
+
+ if (!pci_channel_offline(dev->persist->pdev)) {
+ err = read_vendor_id(dev);
+ /* If PCI can't be accessed to read vendor ID we assume that its
+ * link was disabled and chip was already reset.
+ */
+ if (err)
+ return 0;
+
+ err = mlx4_reset(dev);
+ if (err)
+ mlx4_err(dev, "Fail to reset HCA\n");
+ }
+
+ return err;
+}
+
+static int mlx4_reset_slave(struct mlx4_dev *dev)
+{
+#define COM_CHAN_RST_REQ_OFFSET 0x10
+#define COM_CHAN_RST_ACK_OFFSET 0x08
+
+ u32 comm_flags;
+ u32 rst_req;
+ u32 rst_ack;
+ unsigned long end;
+ struct mlx4_priv *priv = mlx4_priv(dev);
+
+ if (pci_channel_offline(dev->persist->pdev))
+ return 0;
+
+ comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
+ MLX4_COMM_CHAN_FLAGS));
+ if (comm_flags == 0xffffffff) {
+ mlx4_err(dev, "VF reset is not needed\n");
+ return 0;
+ }
+
+ if (!(dev->caps.vf_caps & MLX4_VF_CAP_FLAG_RESET)) {
+ mlx4_err(dev, "VF reset is not supported\n");
+ return -EOPNOTSUPP;
+ }
+
+ rst_req = (comm_flags & (u32)(1 << COM_CHAN_RST_REQ_OFFSET)) >>
+ COM_CHAN_RST_REQ_OFFSET;
+ rst_ack = (comm_flags & (u32)(1 << COM_CHAN_RST_ACK_OFFSET)) >>
+ COM_CHAN_RST_ACK_OFFSET;
+ if (rst_req != rst_ack) {
+ mlx4_err(dev, "Communication channel isn't sync, fail to send reset\n");
+ return -EIO;
+ }
+
+ rst_req ^= 1;
+ mlx4_warn(dev, "VF is sending reset request to Firmware\n");
+ comm_flags = rst_req << COM_CHAN_RST_REQ_OFFSET;
+ __raw_writel((__force u32)cpu_to_be32(comm_flags),
+ (__iomem char *)priv->mfunc.comm + MLX4_COMM_CHAN_FLAGS);
+ /* Make sure that our comm channel write doesn't
+ * get mixed in with writes from another CPU.
+ */
+ mmiowb();
+
+ end = msecs_to_jiffies(MLX4_COMM_TIME) + jiffies;
+ while (time_before(jiffies, end)) {
+ comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
+ MLX4_COMM_CHAN_FLAGS));
+ rst_ack = (comm_flags & (u32)(1 << COM_CHAN_RST_ACK_OFFSET)) >>
+ COM_CHAN_RST_ACK_OFFSET;
+
+ /* Reading rst_req again since the communication channel can
+ * be reset at any time by the PF and all its bits will be
+ * set to zero.
+ */
+ rst_req = (comm_flags & (u32)(1 << COM_CHAN_RST_REQ_OFFSET)) >>
+ COM_CHAN_RST_REQ_OFFSET;
+
+ if (rst_ack == rst_req) {
+ mlx4_warn(dev, "VF Reset succeed\n");
+ return 0;
+ }
+ cond_resched();
+ }
+ mlx4_err(dev, "Fail to send reset over the communication channel\n");
+ return -ETIMEDOUT;
+}
+
+static int mlx4_comm_internal_err(u32 slave_read)
+{
+ return (u32)COMM_CHAN_EVENT_INTERNAL_ERR ==
+ (slave_read & (u32)COMM_CHAN_EVENT_INTERNAL_ERR) ? 1 : 0;
+}
+
+void mlx4_enter_error_state(struct mlx4_dev_persistent *persist)
+{
+ int err;
+ struct mlx4_dev *dev;
+
+ if (!mlx4_internal_err_reset)
+ return;
+
+ mutex_lock(&persist->device_state_mutex);
+ if (persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+ goto out;
+
+ dev = persist->dev;
+ mlx4_err(dev, "device is going to be reset\n");
+ if (mlx4_is_slave(dev))
+ err = mlx4_reset_slave(dev);
+ else
+ err = mlx4_reset_master(dev);
+ BUG_ON(err != 0);
+
+ dev->persist->state |= MLX4_DEVICE_STATE_INTERNAL_ERROR;
+ mlx4_err(dev, "device was reset successfully\n");
+ mutex_unlock(&persist->device_state_mutex);
+
+ /* At that step HW was already reset, now notify clients */
+ mlx4_dispatch_event(dev, MLX4_DEV_EVENT_CATASTROPHIC_ERROR, 0);
+ mlx4_cmd_wake_completions(dev);
+ return;
+
+out:
+ mutex_unlock(&persist->device_state_mutex);
+}
+
+static void mlx4_handle_error_state(struct mlx4_dev_persistent *persist)
+{
+ int err = 0;
+
+ mlx4_enter_error_state(persist);
+ mutex_lock(&persist->interface_state_mutex);
+ if (persist->interface_state & MLX4_INTERFACE_STATE_UP &&
+ !(persist->interface_state & MLX4_INTERFACE_STATE_DELETION)) {
+ err = mlx4_restart_one(persist->pdev);
+ mlx4_info(persist->dev, "mlx4_restart_one was ended, ret=%d\n",
+ err);
+ }
+ mutex_unlock(&persist->interface_state_mutex);
+}