linux-rdma.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Jason Gunthorpe <jgg@ziepe.ca>
To: linux-rdma@vger.kernel.org
Cc: Artemy Kovalyov <artemyko@mellanox.com>,
	Yishai Hadas <yishaih@mellanox.com>
Subject: Re: [PATCH 14/15] RDMA/mlx5: Do not race with mlx5_ib_invalidate_range during create and destroy
Date: Mon, 28 Oct 2019 11:18:44 -0300	[thread overview]
Message-ID: <20191028141844.GA31700@ziepe.ca> (raw)
In-Reply-To: <20191009160934.3143-15-jgg@ziepe.ca>

On Wed, Oct 09, 2019 at 01:09:34PM -0300, Jason Gunthorpe wrote:
> diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
> index 66523313c3e46c..fd2306aff78ad7 100644
> +++ b/drivers/infiniband/hw/mlx5/odp.c
> @@ -144,6 +144,32 @@ void mlx5_odp_populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries,
>  	}
>  }
>  
> +static void dma_fence_odp_mr(struct mlx5_ib_mr *mr)
> +{
> +	struct ib_umem_odp *odp = to_ib_umem_odp(mr->umem);
> +
> +	/* Ensure mlx5_ib_invalidate_range() will not touch the MR any more */
> +	mutex_lock(&odp->umem_mutex);
> +	if (odp->npages) {
> +		/*
> +		 * If not cached then the caller had to do clean_mrs first to
> +		 * fence the mkey.
> +		 */
> +		if (mr->allocated_from_cache) {
> +			mlx5_mr_cache_invalidate(mr);
> +		} else {
> +			/* clean_mr() */
> +			mlx5_core_destroy_mkey(mr->dev->mdev, &mr->mmkey);
> +			WARN_ON(mr->descs);
> +		}
> +		ib_umem_odp_unmap_dma_pages(odp, ib_umem_start(odp),
> +					    ib_umem_end(odp));
> +		WARN_ON(odp->npages);
> +	}
> +	odp->private = NULL;
> +	mutex_unlock(&odp->umem_mutex);

The new mmu notifier debugging catches that mlx5_core_destroy_mkey()
allocates as GFP_KERNEL and cannot be called under umem_mutex, so this
needs to be reworked.

I think to rely on mlx5_mr_cache_invalidate() to fence the DMA, then
destroy out of the lock:

diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 74c45356d54007..f713eb82eeead4 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -151,23 +151,18 @@ static void dma_fence_odp_mr(struct mlx5_ib_mr *mr)
 	/* Ensure mlx5_ib_invalidate_range() will not touch the MR any more */
 	mutex_lock(&odp->umem_mutex);
 	if (odp->npages) {
-		/*
-		 * If not cached then the caller had to do clean_mrs first to
-		 * fence the mkey.
-		 */
-		if (mr->allocated_from_cache) {
-			mlx5_mr_cache_invalidate(mr);
-		} else {
-			/* clean_mr() */
-			mlx5_core_destroy_mkey(mr->dev->mdev, &mr->mmkey);
-			WARN_ON(mr->descs);
-		}
+		mlx5_mr_cache_invalidate(mr);
 		ib_umem_odp_unmap_dma_pages(odp, ib_umem_start(odp),
 					    ib_umem_end(odp));
 		WARN_ON(odp->npages);
 	}
 	odp->private = NULL;
 	mutex_unlock(&odp->umem_mutex);
+
+	if (!mr->allocated_from_cache) {
+		mlx5_core_destroy_mkey(mr->dev->mdev, &mr->mmkey);
+		WARN_ON(mr->descs);
+	}
 }
 
 /*

  reply	other threads:[~2019-10-28 14:18 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-10-09 16:09 [PATCH 00/15] Rework the locking and datastructures for mlx5 implicit ODP Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 01/15] RDMA/mlx5: Use SRCU properly in ODP prefetch Jason Gunthorpe
2019-10-25 19:21   ` Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 02/15] RDMA/mlx5: Split sig_err MR data into its own xarray Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 03/15] RDMA/mlx5: Use a dedicated mkey xarray for ODP Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 04/15] RDMA/mlx5: Delete struct mlx5_priv->mkey_table Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 05/15] RDMA/mlx5: Rework implicit_mr_get_data Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 06/15] RDMA/mlx5: Lift implicit_mr_alloc() into the two routines that call it Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 07/15] RDMA/mlx5: Set the HW IOVA of the child MRs to their place in the tree Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 08/15] RDMA/mlx5: Split implicit handling from pagefault_mr Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 09/15] RDMA/mlx5: Use an xarray for the children of an implicit ODP Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 10/15] RDMA/mlx5: Reduce locking in implicit_mr_get_data() Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 11/15] RDMA/mlx5: Avoid double lookups on the pagefault path Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 12/15] RDMA/mlx5: Rework implicit ODP destroy Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 13/15] RDMA/mlx5: Do not store implicit children in the odp_mkeys xarray Jason Gunthorpe
2019-10-09 16:09 ` [PATCH 14/15] RDMA/mlx5: Do not race with mlx5_ib_invalidate_range during create and destroy Jason Gunthorpe
2019-10-28 14:18   ` Jason Gunthorpe [this message]
2019-10-09 16:09 ` [PATCH 15/15] RDMA/odp: Remove broken debugging call to invalidate_range Jason Gunthorpe
2019-10-28 19:47 ` [PATCH 00/15] Rework the locking and datastructures for mlx5 implicit ODP Jason Gunthorpe

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20191028141844.GA31700@ziepe.ca \
    --to=jgg@ziepe.ca \
    --cc=artemyko@mellanox.com \
    --cc=linux-rdma@vger.kernel.org \
    --cc=yishaih@mellanox.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).