Commit f6bc11e4 authored by Yishai Hadas's avatar Yishai Hadas Committed by David S. Miller

net/mlx4_core: Enhance the catas flow to support device reset

This includes:

- resetting the chip when a fatal error is detected (the current code
  does not do this).

- exposing the ability to enter error state from outside the catas code
  by calling its functionality. (E.g. FW Command timeout, AER error).

- managing a persistent device state. This is needed to sync between
  reset flow cases.
Signed-off-by: default avatarYishai Hadas <yishaih@mellanox.com>
Signed-off-by: default avatarOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent ad9a0bf0
......@@ -48,6 +48,83 @@ MODULE_PARM_DESC(internal_err_reset,
"Reset device on internal errors if non-zero"
" (default 1, in SRIOV mode default is 0)");
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 (!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;
}
void mlx4_enter_error_state(struct mlx4_dev_persistent *persist)
{
int err;
struct mlx4_dev *dev;
if (!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");
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);
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);
err = mlx4_restart_one(persist->pdev);
mlx4_info(persist->dev, "mlx4_restart_one was ended, ret=%d\n", err);
}
static void dump_err_buf(struct mlx4_dev *dev)
{
struct mlx4_priv *priv = mlx4_priv(dev);
......@@ -66,21 +143,22 @@ static void poll_catas(unsigned long dev_ptr)
struct mlx4_priv *priv = mlx4_priv(dev);
if (readl(priv->catas_err.map)) {
/* If the device is off-line, we cannot try to recover it */
if (pci_channel_offline(dev->persist->pdev))
mod_timer(&priv->catas_err.timer,
round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
else {
dump_err_buf(dev);
mlx4_dispatch_event(dev, MLX4_DEV_EVENT_CATASTROPHIC_ERROR, 0);
goto internal_err;
}
if (internal_err_reset)
queue_work(dev->persist->catas_wq,
&dev->persist->catas_work);
if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
mlx4_warn(dev, "Internal error mark was detected on device\n");
goto internal_err;
}
} else
mod_timer(&priv->catas_err.timer,
round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
return;
internal_err:
if (internal_err_reset)
queue_work(dev->persist->catas_wq, &dev->persist->catas_work);
}
static void catas_reset(struct work_struct *work)
......@@ -88,20 +166,8 @@ static void catas_reset(struct work_struct *work)
struct mlx4_dev_persistent *persist =
container_of(work, struct mlx4_dev_persistent,
catas_work);
struct pci_dev *pdev = persist->pdev;
int ret;
/* If the device is off-line, we cannot reset it */
if (pci_channel_offline(pdev))
return;
ret = mlx4_restart_one(pdev);
/* 'priv' now is not valid */
if (ret)
pr_err("mlx4 %s: Reset failed (%d)\n",
pci_name(pdev), ret);
else
mlx4_dbg(persist->dev, "Reset succeeded\n");
mlx4_handle_error_state(persist);
}
void mlx4_start_catas_poll(struct mlx4_dev *dev)
......
......@@ -2624,6 +2624,11 @@ static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
}
}
/* on load remove any previous indication of internal error,
* device is up.
*/
dev->persist->state = MLX4_DEVICE_STATE_UP;
slave_start:
err = mlx4_cmd_init(dev);
if (err) {
......@@ -3108,6 +3113,7 @@ static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
dev->persist->dev = dev;
pci_set_drvdata(pdev, dev->persist);
priv->pci_dev_data = id->driver_data;
mutex_init(&dev->persist->device_state_mutex);
ret = __mlx4_init_one(pdev, id->driver_data, priv);
if (ret) {
......
......@@ -1178,7 +1178,7 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type);
void mlx4_srq_event(struct mlx4_dev *dev, u32 srqn, int event_type);
void mlx4_handle_catas_err(struct mlx4_dev *dev);
void mlx4_enter_error_state(struct mlx4_dev_persistent *persist);
int mlx4_SENSE_PORT(struct mlx4_dev *dev, int port,
enum mlx4_port_type *type);
......
......@@ -411,6 +411,11 @@ enum {
MLX4_EQ_PORT_INFO_MSTR_SM_SL_CHANGE_MASK = 1 << 4,
};
enum {
MLX4_DEVICE_STATE_UP = 1 << 0,
MLX4_DEVICE_STATE_INTERNAL_ERROR = 1 << 1,
};
#define MSTR_SM_CHANGE_MASK (MLX4_EQ_PORT_INFO_MSTR_SM_SL_CHANGE_MASK | \
MLX4_EQ_PORT_INFO_MSTR_SM_LID_CHANGE_MASK)
......@@ -753,6 +758,8 @@ struct mlx4_dev_persistent {
enum mlx4_port_type curr_port_poss_type[MLX4_MAX_PORTS + 1];
struct work_struct catas_work;
struct workqueue_struct *catas_wq;
struct mutex device_state_mutex; /* protect HW state */
u8 state;
};
struct mlx4_dev {
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment