Re: [PATCH rdma-next 02/12] RDMA/odp: Iterate over the whole rbtree directly

From: Jason Gunthorpe <jgg@ziepe.ca>
To: Leon Romanovsky <leon@kernel.org>
Cc: Doug Ledford <dledford@redhat.com>,
	Leon Romanovsky <leonro@mellanox.com>,
	RDMA mailing list <linux-rdma@vger.kernel.org>,
	Guy Levi <guyle@mellanox.com>, Moni Shoua <monis@mellanox.com>
Subject: Re: [PATCH rdma-next 02/12] RDMA/odp: Iterate over the whole rbtree directly
Date: Wed, 21 Aug 2019 14:15:02 -0300	[thread overview]
Message-ID: <20190821171502.GA23022@ziepe.ca> (raw)
In-Reply-To: <20190819111710.18440-3-leon@kernel.org>

On Mon, Aug 19, 2019 at 02:17:00PM +0300, Leon Romanovsky wrote:
> From: Jason Gunthorpe <jgg@mellanox.com>
> 
> Instead of intersecting a full interval, just iterate over every element
> directly. This is faster and clearer.
> 
> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
>  drivers/infiniband/core/umem_odp.c | 51 ++++++++++++++++--------------
>  drivers/infiniband/hw/mlx5/odp.c   | 41 +++++++++++-------------
>  2 files changed, 47 insertions(+), 45 deletions(-)
> 
> diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c
> index 8358eb8e3a26..b9bebef00a33 100644
> +++ b/drivers/infiniband/core/umem_odp.c
> @@ -72,35 +72,41 @@ static void ib_umem_notifier_end_account(struct ib_umem_odp *umem_odp)
>  	mutex_unlock(&umem_odp->umem_mutex);
>  }
>  
> -static int ib_umem_notifier_release_trampoline(struct ib_umem_odp *umem_odp,
> -					       u64 start, u64 end, void *cookie)
> -{
> -	/*
> -	 * Increase the number of notifiers running, to
> -	 * prevent any further fault handling on this MR.
> -	 */
> -	ib_umem_notifier_start_account(umem_odp);
> -	umem_odp->dying = 1;

This patch was not applied on top of the commit noted in the cover
letter

> -	/* Make sure that the fact the umem is dying is out before we release
> -	 * all pending page faults. */
> -	smp_wmb();
> -	complete_all(&umem_odp->notifier_completion);
> -	umem_odp->umem.context->invalidate_range(
> -		umem_odp, ib_umem_start(umem_odp), ib_umem_end(umem_odp));
> -	return 0;
> -}
> -
>  static void ib_umem_notifier_release(struct mmu_notifier *mn,
>  				     struct mm_struct *mm)
>  {
>  	struct ib_ucontext_per_mm *per_mm =
>  		container_of(mn, struct ib_ucontext_per_mm, mn);
> +	struct rb_node *node;
>  
>  	down_read(&per_mm->umem_rwsem);
> -	if (per_mm->active)
> -		rbt_ib_umem_for_each_in_range(
> -			&per_mm->umem_tree, 0, ULLONG_MAX,
> -			ib_umem_notifier_release_trampoline, true, NULL);
> +	if (!per_mm->active)
> +		goto out;
> +
> +	for (node = rb_first_cached(&per_mm->umem_tree); node;
> +	     node = rb_next(node)) {
> +		struct ib_umem_odp *umem_odp =
> +			rb_entry(node, struct ib_umem_odp, interval_tree.rb);
> +
> +		/*
> +		 * Increase the number of notifiers running, to prevent any
> +		 * further fault handling on this MR.
> +		 */
> +		ib_umem_notifier_start_account(umem_odp);
> +
> +		umem_odp->dying = 1;

So this ends up as a 'rebasing error'

I fixed it

Jason