From: Dragos Tatulea <dtatulea@nvidia.com>
To: "Michael S . Tsirkin" <mst@redhat.com>,
Jason Wang <jasowang@redhat.com>,
Eugenio Perez Martin <eperezma@redhat.com>,
Si-Wei Liu <si-wei.liu@oracle.com>,
<virtualization@lists.linux.dev>, Gal Pressman <gal@nvidia.com>
Cc: Dragos Tatulea <dtatulea@nvidia.com>, <kvm@vger.kernel.org>,
<linux-kernel@vger.kernel.org>, Parav Pandit <parav@nvidia.com>,
Xuan Zhuo <xuanzhuo@linux.alibaba.com>,
Cosmin Ratiu <cratiu@nvidia.com>
Subject: [PATCH vhost v2 2/7] vdpa/mlx5: Delete direct MKEYs in parallel
Date: Fri, 30 Aug 2024 13:58:33 +0300 [thread overview]
Message-ID: <20240830105838.2666587-4-dtatulea@nvidia.com> (raw)
In-Reply-To: <20240830105838.2666587-2-dtatulea@nvidia.com>
Use the async interface to issue MTT MKEY deletion.
This makes destroy_user_mr() on average 8x times faster.
This number is also dependent on the size of the MR being
deleted.
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Reviewed-by: Cosmin Ratiu <cratiu@nvidia.com>
Acked-by: Eugenio Pérez <eperezma@redhat.com>
---
drivers/vdpa/mlx5/core/mr.c | 64 +++++++++++++++++++++++++++++++++++++
1 file changed, 64 insertions(+)
diff --git a/drivers/vdpa/mlx5/core/mr.c b/drivers/vdpa/mlx5/core/mr.c
index e72fb11e353d..64bcae2bae8a 100644
--- a/drivers/vdpa/mlx5/core/mr.c
+++ b/drivers/vdpa/mlx5/core/mr.c
@@ -55,6 +55,11 @@ struct mlx5_create_mkey_mem {
__be64 mtt[];
};
+struct mlx5_destroy_mkey_mem {
+ u8 out[MLX5_ST_SZ_BYTES(destroy_mkey_out)];
+ u8 in[MLX5_ST_SZ_BYTES(destroy_mkey_in)];
+};
+
static void fill_create_direct_mr(struct mlx5_vdpa_dev *mvdev,
struct mlx5_vdpa_direct_mr *mr,
struct mlx5_create_mkey_mem *mem)
@@ -91,6 +96,17 @@ static void create_direct_mr_end(struct mlx5_vdpa_dev *mvdev,
mr->mr = mlx5_idx_to_mkey(mkey_index);
}
+static void fill_destroy_direct_mr(struct mlx5_vdpa_dev *mvdev,
+ struct mlx5_vdpa_direct_mr *mr,
+ struct mlx5_destroy_mkey_mem *mem)
+{
+ void *in = &mem->in;
+
+ MLX5_SET(destroy_mkey_in, in, uid, mvdev->res.uid);
+ MLX5_SET(destroy_mkey_in, in, opcode, MLX5_CMD_OP_DESTROY_MKEY);
+ MLX5_SET(destroy_mkey_in, in, mkey_index, mlx5_mkey_to_idx(mr->mr));
+}
+
static void destroy_direct_mr(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_direct_mr *mr)
{
if (!mr->mr)
@@ -257,6 +273,53 @@ static int create_direct_keys(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *
return err;
}
+DEFINE_FREE(free_cmds, struct mlx5_vdpa_async_cmd *, kvfree(_T))
+DEFINE_FREE(free_cmd_mem, struct mlx5_destroy_mkey_mem *, kvfree(_T))
+
+static int destroy_direct_keys(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *mr)
+{
+ struct mlx5_destroy_mkey_mem *cmd_mem __free(free_cmd_mem) = NULL;
+ struct mlx5_vdpa_async_cmd *cmds __free(free_cmds) = NULL;
+ struct mlx5_vdpa_direct_mr *dmr;
+ int err = 0;
+ int i = 0;
+
+ cmds = kvcalloc(mr->num_directs, sizeof(*cmds), GFP_KERNEL);
+ cmd_mem = kvcalloc(mr->num_directs, sizeof(*cmd_mem), GFP_KERNEL);
+ if (!cmds || !cmd_mem)
+ return -ENOMEM;
+
+ list_for_each_entry(dmr, &mr->head, list) {
+ cmds[i].out = cmd_mem[i].out;
+ cmds[i].outlen = sizeof(cmd_mem[i].out);
+ cmds[i].in = cmd_mem[i].in;
+ cmds[i].inlen = sizeof(cmd_mem[i].in);
+ fill_destroy_direct_mr(mvdev, dmr, &cmd_mem[i]);
+ i++;
+ }
+
+ err = mlx5_vdpa_exec_async_cmds(mvdev, cmds, mr->num_directs);
+ if (err) {
+
+ mlx5_vdpa_err(mvdev, "error issuing MTT mkey deletion for direct mrs: %d\n", err);
+ return err;
+ }
+
+ i = 0;
+ list_for_each_entry(dmr, &mr->head, list) {
+ struct mlx5_vdpa_async_cmd *cmd = &cmds[i++];
+
+ dmr->mr = 0;
+ if (cmd->err) {
+ err = err ? err : cmd->err;
+ mlx5_vdpa_err(mvdev, "error deleting MTT mkey [0x%llx, 0x%llx]: %d\n",
+ dmr->start, dmr->end, cmd->err);
+ }
+ }
+
+ return err;
+}
+
static int create_indirect_key(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *mr)
{
int inlen;
@@ -565,6 +628,7 @@ static void destroy_user_mr(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *mr
struct mlx5_vdpa_direct_mr *n;
destroy_indirect_key(mvdev, mr);
+ destroy_direct_keys(mvdev, mr);
list_for_each_entry_safe_reverse(dmr, n, &mr->head, list) {
list_del_init(&dmr->list);
unmap_direct_mr(mvdev, dmr);
--
2.45.1
next prev parent reply other threads:[~2024-08-30 10:59 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-08-30 10:58 [PATCH vhost v2 0/7] vdpa/mlx5: Optimze MKEY operations Dragos Tatulea
2024-08-30 10:58 ` [PATCH vhost v2 1/7] vdpa/mlx5: Create direct MKEYs in parallel Dragos Tatulea
2024-08-30 10:58 ` Dragos Tatulea [this message]
2024-08-30 10:58 ` [PATCH vhost v2 3/7] vdpa/mlx5: Rename function Dragos Tatulea
2024-08-30 10:58 ` [PATCH vhost v2 4/7] vdpa/mlx5: Extract mr members in own resource struct Dragos Tatulea
2024-08-30 10:58 ` [PATCH vhost v2 5/7] vdpa/mlx5: Rename mr_mtx -> lock Dragos Tatulea
2024-08-30 10:58 ` [PATCH vhost v2 6/7] vdpa/mlx5: Introduce init/destroy for MR resources Dragos Tatulea
2024-08-30 10:58 ` [PATCH vhost v2 7/7] vdpa/mlx5: Postpone MR deletion Dragos Tatulea
2024-09-09 9:30 ` [PATCH vhost v2 0/7] vdpa/mlx5: Optimze MKEY operations Dragos Tatulea
2024-09-11 8:02 ` Eugenio Perez Martin
2024-09-11 17:05 ` Dragos Tatulea
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=20240830105838.2666587-4-dtatulea@nvidia.com \
--to=dtatulea@nvidia.com \
--cc=cratiu@nvidia.com \
--cc=eperezma@redhat.com \
--cc=gal@nvidia.com \
--cc=jasowang@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mst@redhat.com \
--cc=parav@nvidia.com \
--cc=si-wei.liu@oracle.com \
--cc=virtualization@lists.linux.dev \
--cc=xuanzhuo@linux.alibaba.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
all inboxes | Powered by JetHome®