net/mlx5: Fix incorrect page count when in internal error
authorDaniel Jurgens <danielj@mellanox.com>
Thu, 30 Jun 2016 14:34:40 +0000 (17:34 +0300)
committerDavid S. Miller <davem@davemloft.net>
Fri, 1 Jul 2016 10:12:03 +0000 (06:12 -0400)
Change page cleanup flow when in internal error to properly decrement
the page counts when reclaiming pages.  The prevents timing out waiting
for extra pages that were actually cleaned up previously.

fixes: 89d44f0a6c73 ('net/mlx5_core: Add pci error handlers to mlx5_core driver')
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c

index 9eeee0545f1cf294ecddb7bb0becc873df8906da..32dea3524cee3ed4d984eb5aafc7a06abdbd713b 100644 (file)
@@ -345,7 +345,6 @@ retry:
                               func_id, npages, err);
                goto out_4k;
        }
-       dev->priv.fw_pages += npages;
 
        err = mlx5_cmd_status_to_err(&out.hdr);
        if (err) {
@@ -373,6 +372,33 @@ out_free:
        return err;
 }
 
+static int reclaim_pages_cmd(struct mlx5_core_dev *dev,
+                            struct mlx5_manage_pages_inbox *in, int in_size,
+                            struct mlx5_manage_pages_outbox *out, int out_size)
+{
+       struct fw_page *fwp;
+       struct rb_node *p;
+       u32 npages;
+       u32 i = 0;
+
+       if (dev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR)
+               return mlx5_cmd_exec_check_status(dev, (u32 *)in, in_size,
+                                                 (u32 *)out, out_size);
+
+       npages = be32_to_cpu(in->num_entries);
+
+       p = rb_first(&dev->priv.page_root);
+       while (p && i < npages) {
+               fwp = rb_entry(p, struct fw_page, rb_node);
+               out->pas[i] = cpu_to_be64(fwp->addr);
+               p = rb_next(p);
+               i++;
+       }
+
+       out->num_entries = cpu_to_be32(i);
+       return 0;
+}
+
 static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
                         int *nclaimed)
 {
@@ -398,15 +424,9 @@ static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
        in.func_id = cpu_to_be16(func_id);
        in.num_entries = cpu_to_be32(npages);
        mlx5_core_dbg(dev, "npages %d, outlen %d\n", npages, outlen);
-       err = mlx5_cmd_exec(dev, &in, sizeof(in), out, outlen);
+       err = reclaim_pages_cmd(dev, &in, sizeof(in), out, outlen);
        if (err) {
-               mlx5_core_err(dev, "failed reclaiming pages\n");
-               goto out_free;
-       }
-       dev->priv.fw_pages -= npages;
-
-       if (out->hdr.status) {
-               err = mlx5_cmd_status_to_err(&out->hdr);
+               mlx5_core_err(dev, "failed reclaiming pages: err %d\n", err);
                goto out_free;
        }
 
@@ -417,13 +437,15 @@ static int reclaim_pages(struct mlx5_core_dev *dev, u32 func_id, int npages,
                err = -EINVAL;
                goto out_free;
        }
-       if (nclaimed)
-               *nclaimed = num_claimed;
 
        for (i = 0; i < num_claimed; i++) {
                addr = be64_to_cpu(out->pas[i]);
                free_4k(dev, addr);
        }
+
+       if (nclaimed)
+               *nclaimed = num_claimed;
+
        dev->priv.fw_pages -= num_claimed;
        if (func_id)
                dev->priv.vfs_pages -= num_claimed;
@@ -514,14 +536,10 @@ int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
                p = rb_first(&dev->priv.page_root);
                if (p) {
                        fwp = rb_entry(p, struct fw_page, rb_node);
-                       if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
-                               free_4k(dev, fwp->addr);
-                               nclaimed = 1;
-                       } else {
-                               err = reclaim_pages(dev, fwp->func_id,
-                                                   optimal_reclaimed_pages(),
-                                                   &nclaimed);
-                       }
+                       err = reclaim_pages(dev, fwp->func_id,
+                                           optimal_reclaimed_pages(),
+                                           &nclaimed);
+
                        if (err) {
                                mlx5_core_warn(dev, "failed reclaiming pages (%d)\n",
                                               err);
@@ -536,6 +554,13 @@ int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev)
                }
        } while (p);
 
+       WARN(dev->priv.fw_pages,
+            "FW pages counter is %d after reclaiming all pages\n",
+            dev->priv.fw_pages);
+       WARN(dev->priv.vfs_pages,
+            "VFs FW pages counter is %d after reclaiming all pages\n",
+            dev->priv.vfs_pages);
+
        return 0;
 }