Reset the device when an internal error is detected.
Also, detect errors by polling the error buffer rather than using
interrupts. This is more robust and doesn't depend on MSI-X. Remove
the old interrupt handler entirely, since we don't want to support two
mechanisms for detecting internal errors.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
* SOFTWARE.
*/
+#include <linux/workqueue.h>
+
#include "mlx4.h"
-void mlx4_handle_catas_err(struct mlx4_dev *dev)
+enum {
+ MLX4_CATAS_POLL_INTERVAL = 5 * HZ,
+};
+
+static DEFINE_SPINLOCK(catas_lock);
+
+static LIST_HEAD(catas_list);
+static struct workqueue_struct *catas_wq;
+static struct work_struct catas_work;
+
+static int internal_err_reset = 1;
+module_param(internal_err_reset, int, 0644);
+MODULE_PARM_DESC(internal_err_reset,
+ "Reset device on internal errors if non-zero (default 1)");
+
+static void dump_err_buf(struct mlx4_dev *dev)
{
struct mlx4_priv *priv = mlx4_priv(dev);
int i;
- mlx4_err(dev, "Catastrophic error detected:\n");
+ mlx4_err(dev, "Internal error detected:\n");
for (i = 0; i < priv->fw.catas_size; ++i)
mlx4_err(dev, " buf[%02x]: %08x\n",
i, swab32(readl(priv->catas_err.map + i)));
+}
- mlx4_dispatch_event(dev, MLX4_EVENT_TYPE_LOCAL_CATAS_ERROR, 0, 0);
+static void poll_catas(unsigned long dev_ptr)
+{
+ struct mlx4_dev *dev = (struct mlx4_dev *) dev_ptr;
+ struct mlx4_priv *priv = mlx4_priv(dev);
+
+ if (readl(priv->catas_err.map)) {
+ dump_err_buf(dev);
+
+ mlx4_dispatch_event(dev, MLX4_EVENT_TYPE_LOCAL_CATAS_ERROR, 0, 0);
+
+ if (internal_err_reset) {
+ spin_lock(&catas_lock);
+ list_add(&priv->catas_err.list, &catas_list);
+ spin_unlock(&catas_lock);
+
+ queue_work(catas_wq, &catas_work);
+ }
+ } else
+ mod_timer(&priv->catas_err.timer,
+ round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
}
-void mlx4_map_catas_buf(struct mlx4_dev *dev)
+static void catas_reset(struct work_struct *work)
+{
+ struct mlx4_priv *priv, *tmppriv;
+ struct mlx4_dev *dev;
+
+ LIST_HEAD(tlist);
+ int ret;
+
+ spin_lock_irq(&catas_lock);
+ list_splice_init(&catas_list, &tlist);
+ spin_unlock_irq(&catas_lock);
+
+ list_for_each_entry_safe(priv, tmppriv, &tlist, catas_err.list) {
+ ret = mlx4_restart_one(priv->dev.pdev);
+ dev = &priv->dev;
+ if (ret)
+ mlx4_err(dev, "Reset failed (%d)\n", ret);
+ else
+ mlx4_dbg(dev, "Reset succeeded\n");
+ }
+}
+
+void mlx4_start_catas_poll(struct mlx4_dev *dev)
{
struct mlx4_priv *priv = mlx4_priv(dev);
unsigned long addr;
+ INIT_LIST_HEAD(&priv->catas_err.list);
+ init_timer(&priv->catas_err.timer);
+ priv->catas_err.map = NULL;
+
addr = pci_resource_start(dev->pdev, priv->fw.catas_bar) +
priv->fw.catas_offset;
priv->catas_err.map = ioremap(addr, priv->fw.catas_size * 4);
- if (!priv->catas_err.map)
- mlx4_warn(dev, "Failed to map catastrophic error buffer at 0x%lx\n",
+ if (!priv->catas_err.map) {
+ mlx4_warn(dev, "Failed to map internal error buffer at 0x%lx\n",
addr);
+ return;
+ }
+ priv->catas_err.timer.data = (unsigned long) dev;
+ priv->catas_err.timer.function = poll_catas;
+ priv->catas_err.timer.expires =
+ round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL);
+ add_timer(&priv->catas_err.timer);
}
-void mlx4_unmap_catas_buf(struct mlx4_dev *dev)
+void mlx4_stop_catas_poll(struct mlx4_dev *dev)
{
struct mlx4_priv *priv = mlx4_priv(dev);
+ del_timer_sync(&priv->catas_err.timer);
+
if (priv->catas_err.map)
iounmap(priv->catas_err.map);
+
+ spin_lock_irq(&catas_lock);
+ list_del(&priv->catas_err.list);
+ spin_unlock_irq(&catas_lock);
+}
+
+int __init mlx4_catas_init(void)
+{
+ INIT_WORK(&catas_work, catas_reset);
+
+ catas_wq = create_singlethread_workqueue("mlx4_err");
+ if (!catas_wq)
+ return -ENOMEM;
+
+ return 0;
+}
+
+void mlx4_catas_cleanup(void)
+{
+ destroy_workqueue(catas_wq);
}
(1ull << MLX4_EVENT_TYPE_PATH_MIG_FAILED) | \
(1ull << MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
(1ull << MLX4_EVENT_TYPE_WQ_ACCESS_ERROR) | \
- (1ull << MLX4_EVENT_TYPE_LOCAL_CATAS_ERROR) | \
(1ull << MLX4_EVENT_TYPE_PORT_CHANGE) | \
(1ull << MLX4_EVENT_TYPE_ECC_DETECT) | \
(1ull << MLX4_EVENT_TYPE_SRQ_CATAS_ERROR) | \
(1ull << MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE) | \
(1ull << MLX4_EVENT_TYPE_SRQ_LIMIT) | \
(1ull << MLX4_EVENT_TYPE_CMD))
-#define MLX4_CATAS_EVENT_MASK (1ull << MLX4_EVENT_TYPE_LOCAL_CATAS_ERROR)
struct mlx4_eqe {
u8 reserved1;
writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
- for (i = 0; i < MLX4_EQ_CATAS; ++i)
+ for (i = 0; i < MLX4_NUM_EQ; ++i)
work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
return IRQ_RETVAL(work);
return IRQ_HANDLED;
}
-static irqreturn_t mlx4_catas_interrupt(int irq, void *dev_ptr)
-{
- mlx4_handle_catas_err(dev_ptr);
-
- /* MSI-X vectors always belong to us */
- return IRQ_HANDLED;
-}
-
static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
int eq_num)
{
if (eq_table->have_irq)
free_irq(dev->pdev->irq, dev);
- for (i = 0; i < MLX4_EQ_CATAS; ++i)
+ for (i = 0; i < MLX4_NUM_EQ; ++i)
if (eq_table->eq[i].have_irq)
free_irq(eq_table->eq[i].irq, eq_table->eq + i);
- if (eq_table->eq[MLX4_EQ_CATAS].have_irq)
- free_irq(eq_table->eq[MLX4_EQ_CATAS].irq, dev);
}
static int __devinit mlx4_map_clr_int(struct mlx4_dev *dev)
if (dev->flags & MLX4_FLAG_MSI_X) {
static const char *eq_name[] = {
[MLX4_EQ_COMP] = DRV_NAME " (comp)",
- [MLX4_EQ_ASYNC] = DRV_NAME " (async)",
- [MLX4_EQ_CATAS] = DRV_NAME " (catas)"
+ [MLX4_EQ_ASYNC] = DRV_NAME " (async)"
};
- err = mlx4_create_eq(dev, 1, MLX4_EQ_CATAS,
- &priv->eq_table.eq[MLX4_EQ_CATAS]);
- if (err)
- goto err_out_async;
-
- for (i = 0; i < MLX4_EQ_CATAS; ++i) {
+ for (i = 0; i < MLX4_NUM_EQ; ++i) {
err = request_irq(priv->eq_table.eq[i].irq,
mlx4_msi_x_interrupt,
0, eq_name[i], priv->eq_table.eq + i);
if (err)
- goto err_out_catas;
+ goto err_out_async;
priv->eq_table.eq[i].have_irq = 1;
}
- err = request_irq(priv->eq_table.eq[MLX4_EQ_CATAS].irq,
- mlx4_catas_interrupt, 0,
- eq_name[MLX4_EQ_CATAS], dev);
- if (err)
- goto err_out_catas;
-
- priv->eq_table.eq[MLX4_EQ_CATAS].have_irq = 1;
} else {
err = request_irq(dev->pdev->irq, mlx4_interrupt,
IRQF_SHARED, DRV_NAME, dev);
mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
priv->eq_table.eq[MLX4_EQ_ASYNC].eqn, err);
- for (i = 0; i < MLX4_EQ_CATAS; ++i)
+ for (i = 0; i < MLX4_NUM_EQ; ++i)
eq_set_ci(&priv->eq_table.eq[i], 1);
- if (dev->flags & MLX4_FLAG_MSI_X) {
- err = mlx4_MAP_EQ(dev, MLX4_CATAS_EVENT_MASK, 0,
- priv->eq_table.eq[MLX4_EQ_CATAS].eqn);
- if (err)
- mlx4_warn(dev, "MAP_EQ for catas EQ %d failed (%d)\n",
- priv->eq_table.eq[MLX4_EQ_CATAS].eqn, err);
- }
-
return 0;
-err_out_catas:
- mlx4_free_eq(dev, &priv->eq_table.eq[MLX4_EQ_CATAS]);
-
err_out_async:
mlx4_free_eq(dev, &priv->eq_table.eq[MLX4_EQ_ASYNC]);
struct mlx4_priv *priv = mlx4_priv(dev);
int i;
- if (dev->flags & MLX4_FLAG_MSI_X)
- mlx4_MAP_EQ(dev, MLX4_CATAS_EVENT_MASK, 1,
- priv->eq_table.eq[MLX4_EQ_CATAS].eqn);
-
mlx4_MAP_EQ(dev, MLX4_ASYNC_EVENT_MASK, 1,
priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
mlx4_free_irqs(dev);
- for (i = 0; i < MLX4_EQ_CATAS; ++i)
+ for (i = 0; i < MLX4_NUM_EQ; ++i)
mlx4_free_eq(dev, &priv->eq_table.eq[i]);
- if (dev->flags & MLX4_FLAG_MSI_X)
- mlx4_free_eq(dev, &priv->eq_table.eq[MLX4_EQ_CATAS]);
mlx4_unmap_clr_int(dev);
mlx4_add_device(intf, priv);
mutex_unlock(&intf_mutex);
+ mlx4_start_catas_poll(dev);
return 0;
}
struct mlx4_priv *priv = mlx4_priv(dev);
struct mlx4_interface *intf;
+ mlx4_stop_catas_poll(dev);
mutex_lock(&intf_mutex);
list_for_each_entry(intf, &intf_list, list)
goto err_pd_table_free;
}
- mlx4_map_catas_buf(dev);
-
err = mlx4_init_eq_table(dev);
if (err) {
mlx4_err(dev, "Failed to initialize "
"event queue table, aborting.\n");
- goto err_catas_buf;
+ goto err_mr_table_free;
}
err = mlx4_cmd_use_events(dev);
err_eq_table_free:
mlx4_cleanup_eq_table(dev);
-err_catas_buf:
- mlx4_unmap_catas_buf(dev);
+err_mr_table_free:
mlx4_cleanup_mr_table(dev);
err_pd_table_free:
mlx4_cleanup_cq_table(dev);
mlx4_cmd_use_polling(dev);
mlx4_cleanup_eq_table(dev);
-
- mlx4_unmap_catas_buf(dev);
-
mlx4_cleanup_mr_table(dev);
mlx4_cleanup_pd_table(dev);
mlx4_cleanup_uar_table(dev);
mlx4_cleanup_cq_table(dev);
mlx4_cmd_use_polling(dev);
mlx4_cleanup_eq_table(dev);
-
- mlx4_unmap_catas_buf(dev);
-
mlx4_cleanup_mr_table(dev);
mlx4_cleanup_pd_table(dev);
}
}
+int mlx4_restart_one(struct pci_dev *pdev)
+{
+ mlx4_remove_one(pdev);
+ return mlx4_init_one(pdev, NULL);
+}
+
static struct pci_device_id mlx4_pci_table[] = {
{ PCI_VDEVICE(MELLANOX, 0x6340) }, /* MT25408 "Hermon" SDR */
{ PCI_VDEVICE(MELLANOX, 0x634a) }, /* MT25408 "Hermon" DDR */
{
int ret;
+ ret = mlx4_catas_init();
+ if (ret)
+ return ret;
+
ret = pci_register_driver(&mlx4_driver);
return ret < 0 ? ret : 0;
}
static void __exit mlx4_cleanup(void)
{
pci_unregister_driver(&mlx4_driver);
+ mlx4_catas_cleanup();
}
module_init(mlx4_init);
#include <linux/mutex.h>
#include <linux/radix-tree.h>
+#include <linux/timer.h>
#include <linux/mlx4/device.h>
#include <linux/mlx4/doorbell.h>
enum {
MLX4_EQ_ASYNC,
MLX4_EQ_COMP,
- MLX4_EQ_CATAS,
MLX4_NUM_EQ
};
struct mlx4_catas_err {
u32 __iomem *map;
- int size;
+ struct timer_list timer;
+ struct list_head list;
};
struct mlx4_priv {
void mlx4_cleanup_srq_table(struct mlx4_dev *dev);
void mlx4_cleanup_mcg_table(struct mlx4_dev *dev);
-void mlx4_map_catas_buf(struct mlx4_dev *dev);
-void mlx4_unmap_catas_buf(struct mlx4_dev *dev);
-
+void mlx4_start_catas_poll(struct mlx4_dev *dev);
+void mlx4_stop_catas_poll(struct mlx4_dev *dev);
+int mlx4_catas_init(void);
+void mlx4_catas_cleanup(void);
+int mlx4_restart_one(struct pci_dev *pdev);
int mlx4_register_device(struct mlx4_dev *dev);
void mlx4_unregister_device(struct mlx4_dev *dev);
void mlx4_dispatch_event(struct mlx4_dev *dev, enum mlx4_event type,