On Thu, Sep 28, 2023 at 11:38:51AM +1000, Stephen Rothwell wrote:
> Hi all,
>
> Today's linux-next merge of the rdma tree got a conflict in:
>
> drivers/infiniband/hw/mlx5/mr.c
>
> between commit:
>
> 374012b00457 ("RDMA/mlx5: Fix mkey cache possible deadlock on cleanup")
>
> from the rdma-fixes tree and commit:
>
> 57e7071683ef ("RDMA/mlx5: Implement mkeys management via LIFO queue")
>
> from the rdma tree.
>
> I fixed it up (see below) and can carry the fix as necessary. This
> is now fixed as far as linux-next is concerned, but any non trivial
> conflicts should be mentioned to your upstream maintainer when your tree
> is submitted for merging. You may also want to consider cooperating
> with the maintainer of the conflicting tree to minimise any particularly
> complex conflicts.
>
> --
> Cheers,
> Stephen Rothwell
>
> diff --cc drivers/infiniband/hw/mlx5/mr.c
> index 433f96459246,b0fa2d644973..000000000000
> --- a/drivers/infiniband/hw/mlx5/mr.c
> +++ b/drivers/infiniband/hw/mlx5/mr.c
> @@@ -1025,21 -998,15 +999,21 @@@ void mlx5_mkey_cache_cleanup(struct mlx
> if (!dev->cache.wq)
> return;
>
> - cancel_delayed_work_sync(&dev->cache.remove_ent_dwork);
> mutex_lock(&dev->cache.rb_lock);
> + dev->cache.disable = true;
> for (node = rb_first(root); node; node = rb_next(node)) {
> ent = rb_entry(node, struct mlx5_cache_ent, node);
> - xa_lock_irq(&ent->mkeys);
> + spin_lock_irq(&ent->mkeys_queue.lock);
> ent->disabled = true;
> - xa_unlock_irq(&ent->mkeys);
> + spin_unlock_irq(&ent->mkeys_queue.lock);
> - cancel_delayed_work_sync(&ent->dwork);
> }
> + mutex_unlock(&dev->cache.rb_lock);
> +
> + /*
> + * After all entries are disabled and will not reschedule on WQ,
> + * flush it and all async commands.
> + */
> + flush_workqueue(dev->cache.wq);
>
> mlx5_mkey_cache_debugfs_cleanup(dev);
> mlx5_cmd_cleanup_async_ctx(&dev->async_ctx);
Thanks for the fix.
