* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ivan Khoronzhuk @ 2019-06-30 16:20 UTC (permalink / raw)
To: Ilias Apalodimas
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <1561785805-21647-4-git-send-email-ilias.apalodimas@linaro.org>
On Sat, Jun 29, 2019 at 08:23:25AM +0300, Ilias Apalodimas wrote:
[...]
>+
>+static int netsec_xdp(struct net_device *ndev, struct netdev_bpf *xdp)
>+{
>+ struct netsec_priv *priv = netdev_priv(ndev);
>+
>+ switch (xdp->command) {
>+ case XDP_SETUP_PROG:
>+ return netsec_xdp_setup(priv, xdp->prog, xdp->extack);
>+ case XDP_QUERY_PROG:
>+ xdp->prog_id = priv->xdp_prog ? priv->xdp_prog->aux->id : 0;
xdp_attachment family to save bpf flags?
>+ return 0;
>+ default:
>+ return -EINVAL;
>+ }
>+}
>+
> static const struct net_device_ops netsec_netdev_ops = {
> .ndo_init = netsec_netdev_init,
> .ndo_uninit = netsec_netdev_uninit,
>@@ -1537,6 +1842,8 @@ static const struct net_device_ops netsec_netdev_ops = {
> .ndo_set_mac_address = eth_mac_addr,
> .ndo_validate_addr = eth_validate_addr,
> .ndo_do_ioctl = netsec_netdev_ioctl,
>+ .ndo_xdp_xmit = netsec_xdp_xmit,
>+ .ndo_bpf = netsec_xdp,
> };
>
> static int netsec_of_probe(struct platform_device *pdev,
>--
>2.20.1
>
--
Regards,
Ivan Khoronzhuk
^ permalink raw reply
* [PATCH rdma-next v2 00/13] DEVX asynchronous events
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
From: Leon Romanovsky <leonro@mellanox.com>
Changelog:
v1 -> v2:
* Added Saeed's ack to net patches
* Patch #2:
* Fix to gather user asynchronous events on top of kernel events.
* Patch #7:
* Fix obj_id to be 32 bits.
* Patch #8:
* Inline async_event_queue applicable fields into devx_async_event_file.
* Move to use bitfields in few places rather than flags.
* Shorten name of UAPI attribute.
* Patch #10:
* Use explicitly 'struct file *' instead of void *
* Store struct devx_async_event_file * instead of uobj * on the subscription.
* Drop 'is_obj_related' and use list_empty instead.
* Drop the temp arrays as part of the subscription API and move to simpler logic.
* Revise devx_cleanup_subscription() to be success oriented without
the is_close flag.
* Leave key level 1 in the tree upon bad flow to prevent a race with IRQ flow.
* Fix some styling notes.
* Patch #11:
* Use rcu read lock also for the un-affiliated event flow.
* Improve locking scheme as part of read events.
* Return -EIO as soon as destroyed occurred.
* Use a better errno as part read event failure when the buffer size
was too small.
* Upon hot unplug call wake_up_interruptible() unconditionally.
* Use eqe->data for affiliated events header.
* Fix some styling notes.
* Patch #12:
* Use rcu read lock also for the first XA layer.
* Patch #13:
* A new patch to clean up mdev usage from devx code, it can be accessed
from ib_dev now.
v0 -> v1:
* Fix the unbind / hot unplug flows to work properly.
* Fix Ref count handling on the eventfd mode in some flow.
* Rebased to latest rdma-next
Thanks
------------------------------------------------------------------------------------
From Yishai:
This series enables RDMA applications that use the DEVX interface to
subscribe and read device asynchronous events.
The solution is designed to allow extension of events in the future
without need to perform any changes in the driver code.
To enable that few changes had been done in mlx5_core, it includes:
* Reading device event capabilities that are user related
(affiliated and un-affiliated) and set the matching mask upon
creating the matching EQ.
* Enable DEVX/mlx5_ib to register for ANY event instead of the option to
get some hard-coded ones.
* Enable DEVX/mlx5_ib to get the device raw data for CQ completion events.
* Enhance mlx5_core_create/destroy CQ to enable DEVX using them so that CQ
events will be reported as well.
In mlx5_ib layer the below changes were done:
* A new DEVX API was introduced to allocate an event channel by using
the uverbs FD object type.
* Implement the FD channel operations to enable read/poo/close over it.
* A new DEVX API was introduced to subscribe for specific events over an
event channel.
* Manage an internal data structure over XA(s) to subscribe/dispatch events
over the different event channels.
* Use from DEVX the mlx5_core APIs to create/destroy a CQ to be able to
get its relevant events.
Yishai
Yishai Hadas (13):
net/mlx5: Fix mlx5_core_destroy_cq() error flow
net/mlx5: Use event mask based on device capabilities
net/mlx5: Expose the API to register for ANY event
net/mlx5: mlx5_core_create_cq() enhancements
net/mlx5: Report a CQ error event only when a handler was set
net/mlx5: Report EQE data upon CQ completion
net/mlx5: Expose device definitions for object events
IB/mlx5: Introduce MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD
IB/mlx5: Register DEVX with mlx5_core to get async events
IB/mlx5: Enable subscription for device events over DEVX
IB/mlx5: Implement DEVX dispatching event
IB/mlx5: Add DEVX support for CQ events
IB/mlx5: DEVX cleanup mdev
drivers/infiniband/hw/mlx5/cq.c | 5 +-
drivers/infiniband/hw/mlx5/devx.c | 1029 ++++++++++++++++-
drivers/infiniband/hw/mlx5/main.c | 10 +-
drivers/infiniband/hw/mlx5/mlx5_ib.h | 12 +
drivers/infiniband/hw/mlx5/odp.c | 2 +-
drivers/infiniband/hw/mlx5/qp.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/cq.c | 21 +-
drivers/net/ethernet/mellanox/mlx5/core/en.h | 2 +-
.../net/ethernet/mellanox/mlx5/core/en_main.c | 3 +-
.../net/ethernet/mellanox/mlx5/core/en_txrx.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 63 +-
.../ethernet/mellanox/mlx5/core/fpga/conn.c | 6 +-
drivers/net/ethernet/mellanox/mlx5/core/fw.c | 6 +
.../net/ethernet/mellanox/mlx5/core/lib/eq.h | 5 +-
include/linux/mlx5/cq.h | 6 +-
include/linux/mlx5/device.h | 6 +-
include/linux/mlx5/driver.h | 2 +
include/linux/mlx5/eq.h | 2 +-
include/linux/mlx5/mlx5_ifc.h | 34 +-
include/uapi/rdma/mlx5_user_ioctl_cmds.h | 19 +
include/uapi/rdma/mlx5_user_ioctl_verbs.h | 9 +
21 files changed, 1178 insertions(+), 68 deletions(-)
--
2.20.1
^ permalink raw reply
* [PATCH mlx5-next v2 01/13] net/mlx5: Fix mlx5_core_destroy_cq() error flow
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
The firmware command to destroy a CQ might fail when the object is
referenced by other object and the ref count is managed by the firmware.
To enable a second successful destruction post the first failure need to
change mlx5_eq_del_cq() to be a void function.
As an error in mlx5_eq_del_cq() is quite fatal from the option to
recover, a debug message inside it should be good enougth and it was
changed to be void.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/cq.c | 9 ++-------
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 16 +++++++---------
drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h | 2 +-
3 files changed, 10 insertions(+), 17 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/cq.c b/drivers/net/ethernet/mellanox/mlx5/core/cq.c
index 713a17ee3751..703d88332bc6 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/cq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/cq.c
@@ -158,13 +158,8 @@ int mlx5_core_destroy_cq(struct mlx5_core_dev *dev, struct mlx5_core_cq *cq)
u32 in[MLX5_ST_SZ_DW(destroy_cq_in)] = {0};
int err;
- err = mlx5_eq_del_cq(mlx5_get_async_eq(dev), cq);
- if (err)
- return err;
-
- err = mlx5_eq_del_cq(&cq->eq->core, cq);
- if (err)
- return err;
+ mlx5_eq_del_cq(mlx5_get_async_eq(dev), cq);
+ mlx5_eq_del_cq(&cq->eq->core, cq);
MLX5_SET(destroy_cq_in, in, opcode, MLX5_CMD_OP_DESTROY_CQ);
MLX5_SET(destroy_cq_in, in, cqn, cq->cqn);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eq.c b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
index 58fff2f39b38..8000d2a4a7e2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
@@ -389,7 +389,7 @@ int mlx5_eq_add_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq)
return err;
}
-int mlx5_eq_del_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq)
+void mlx5_eq_del_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq)
{
struct mlx5_cq_table *table = &eq->cq_table;
struct mlx5_core_cq *tmp;
@@ -399,16 +399,14 @@ int mlx5_eq_del_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq)
spin_unlock(&table->lock);
if (!tmp) {
- mlx5_core_warn(eq->dev, "cq 0x%x not found in eq 0x%x tree\n", eq->eqn, cq->cqn);
- return -ENOENT;
+ mlx5_core_dbg(eq->dev, "cq 0x%x not found in eq 0x%x tree\n",
+ eq->eqn, cq->cqn);
+ return;
}
- if (tmp != cq) {
- mlx5_core_warn(eq->dev, "corruption on cqn 0x%x in eq 0x%x\n", eq->eqn, cq->cqn);
- return -EINVAL;
- }
-
- return 0;
+ if (tmp != cq)
+ mlx5_core_dbg(eq->dev, "corruption on cqn 0x%x in eq 0x%x\n",
+ eq->eqn, cq->cqn);
}
int mlx5_eq_table_init(struct mlx5_core_dev *dev)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h b/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
index 24bd991a727e..d826e63d5a17 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
@@ -75,7 +75,7 @@ int mlx5_eq_table_create(struct mlx5_core_dev *dev);
void mlx5_eq_table_destroy(struct mlx5_core_dev *dev);
int mlx5_eq_add_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq);
-int mlx5_eq_del_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq);
+void mlx5_eq_del_cq(struct mlx5_eq *eq, struct mlx5_core_cq *cq);
struct mlx5_eq_comp *mlx5_eqn2comp_eq(struct mlx5_core_dev *dev, int eqn);
struct mlx5_eq *mlx5_get_async_eq(struct mlx5_core_dev *dev);
void mlx5_cq_tasklet_cb(unsigned long data);
--
2.20.1
^ permalink raw reply related
* [PATCH mlx5-next v2 03/13] net/mlx5: Expose the API to register for ANY event
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Expose the API to register for ANY event, mlx5_ib will be able to use
this functionality for its needs.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 2 ++
drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h | 3 ---
include/linux/mlx5/driver.h | 2 ++
3 files changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eq.c b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
index 33f78d4d3724..c634a78d5cdd 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
@@ -949,6 +949,7 @@ int mlx5_eq_notifier_register(struct mlx5_core_dev *dev, struct mlx5_nb *nb)
return atomic_notifier_chain_register(&eqt->nh[nb->event_type], &nb->nb);
}
+EXPORT_SYMBOL(mlx5_eq_notifier_register);
int mlx5_eq_notifier_unregister(struct mlx5_core_dev *dev, struct mlx5_nb *nb)
{
@@ -959,3 +960,4 @@ int mlx5_eq_notifier_unregister(struct mlx5_core_dev *dev, struct mlx5_nb *nb)
return atomic_notifier_chain_unregister(&eqt->nh[nb->event_type], &nb->nb);
}
+EXPORT_SYMBOL(mlx5_eq_notifier_unregister);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h b/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
index d826e63d5a17..3dfab91ae5f2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
@@ -97,7 +97,4 @@ void mlx5_core_eq_free_irqs(struct mlx5_core_dev *dev);
struct cpu_rmap *mlx5_eq_table_get_rmap(struct mlx5_core_dev *dev);
#endif
-int mlx5_eq_notifier_register(struct mlx5_core_dev *dev, struct mlx5_nb *nb);
-int mlx5_eq_notifier_unregister(struct mlx5_core_dev *dev, struct mlx5_nb *nb);
-
#endif
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index d8ab633406c2..3f8b9d8e2070 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -1047,6 +1047,8 @@ int mlx5_register_interface(struct mlx5_interface *intf);
void mlx5_unregister_interface(struct mlx5_interface *intf);
int mlx5_notifier_register(struct mlx5_core_dev *dev, struct notifier_block *nb);
int mlx5_notifier_unregister(struct mlx5_core_dev *dev, struct notifier_block *nb);
+int mlx5_eq_notifier_register(struct mlx5_core_dev *dev, struct mlx5_nb *nb);
+int mlx5_eq_notifier_unregister(struct mlx5_core_dev *dev, struct mlx5_nb *nb);
int mlx5_core_query_vendor_id(struct mlx5_core_dev *mdev, u32 *vendor_id);
--
2.20.1
^ permalink raw reply related
* [PATCH mlx5-next v2 04/13] net/mlx5: mlx5_core_create_cq() enhancements
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Enhance mlx5_core_create_cq() to get the command out buffer from the
callers to let them use the output.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/cq.c | 3 ++-
drivers/net/ethernet/mellanox/mlx5/core/cq.c | 7 +++----
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 3 ++-
drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c | 3 ++-
include/linux/mlx5/cq.h | 2 +-
5 files changed, 10 insertions(+), 8 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index 22230fd7d741..857a554e2c29 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -891,6 +891,7 @@ int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
int entries = attr->cqe;
int vector = attr->comp_vector;
struct mlx5_ib_dev *dev = to_mdev(ibdev);
+ u32 out[MLX5_ST_SZ_DW(create_cq_out)];
struct mlx5_ib_cq *cq = to_mcq(ibcq);
int uninitialized_var(index);
int uninitialized_var(inlen);
@@ -954,7 +955,7 @@ int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
if (cq->create_flags & IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN)
MLX5_SET(cqc, cqc, oi, 1);
- err = mlx5_core_create_cq(dev->mdev, &cq->mcq, cqb, inlen);
+ err = mlx5_core_create_cq(dev->mdev, &cq->mcq, cqb, inlen, out, sizeof(out));
if (err)
goto err_cqb;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/cq.c b/drivers/net/ethernet/mellanox/mlx5/core/cq.c
index 703d88332bc6..1bd4336392a2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/cq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/cq.c
@@ -87,11 +87,10 @@ static void mlx5_add_cq_to_tasklet(struct mlx5_core_cq *cq)
}
int mlx5_core_create_cq(struct mlx5_core_dev *dev, struct mlx5_core_cq *cq,
- u32 *in, int inlen)
+ u32 *in, int inlen, u32 *out, int outlen)
{
int eqn = MLX5_GET(cqc, MLX5_ADDR_OF(create_cq_in, in, cq_context), c_eqn);
u32 dout[MLX5_ST_SZ_DW(destroy_cq_out)];
- u32 out[MLX5_ST_SZ_DW(create_cq_out)];
u32 din[MLX5_ST_SZ_DW(destroy_cq_in)];
struct mlx5_eq_comp *eq;
int err;
@@ -100,9 +99,9 @@ int mlx5_core_create_cq(struct mlx5_core_dev *dev, struct mlx5_core_cq *cq,
if (IS_ERR(eq))
return PTR_ERR(eq);
- memset(out, 0, sizeof(out));
+ memset(out, 0, outlen);
MLX5_SET(create_cq_in, in, opcode, MLX5_CMD_OP_CREATE_CQ);
- err = mlx5_cmd_exec(dev, in, inlen, out, sizeof(out));
+ err = mlx5_cmd_exec(dev, in, inlen, out, outlen);
if (err)
return err;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 1046bd057152..2a1a7dad5d04 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -1518,6 +1518,7 @@ static void mlx5e_free_cq(struct mlx5e_cq *cq)
static int mlx5e_create_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
{
+ u32 out[MLX5_ST_SZ_DW(create_cq_out)];
struct mlx5_core_dev *mdev = cq->mdev;
struct mlx5_core_cq *mcq = &cq->mcq;
@@ -1552,7 +1553,7 @@ static int mlx5e_create_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param)
MLX5_ADAPTER_PAGE_SHIFT);
MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma);
- err = mlx5_core_create_cq(mdev, mcq, in, inlen);
+ err = mlx5_core_create_cq(mdev, mcq, in, inlen, out, sizeof(out));
kvfree(in);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
index ca2296a2f9ee..dc7b9d9f274d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
@@ -429,6 +429,7 @@ static int mlx5_fpga_conn_create_cq(struct mlx5_fpga_conn *conn, int cq_size)
struct mlx5_fpga_device *fdev = conn->fdev;
struct mlx5_core_dev *mdev = fdev->mdev;
u32 temp_cqc[MLX5_ST_SZ_DW(cqc)] = {0};
+ u32 out[MLX5_ST_SZ_DW(create_cq_out)];
struct mlx5_wq_param wqp;
struct mlx5_cqe64 *cqe;
int inlen, err, eqn;
@@ -476,7 +477,7 @@ static int mlx5_fpga_conn_create_cq(struct mlx5_fpga_conn *conn, int cq_size)
pas = (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas);
mlx5_fill_page_frag_array(&conn->cq.wq_ctrl.buf, pas);
- err = mlx5_core_create_cq(mdev, &conn->cq.mcq, in, inlen);
+ err = mlx5_core_create_cq(mdev, &conn->cq.mcq, in, inlen, out, sizeof(out));
kvfree(in);
if (err)
diff --git a/include/linux/mlx5/cq.h b/include/linux/mlx5/cq.h
index 769326ea1d9b..e44157a2b7db 100644
--- a/include/linux/mlx5/cq.h
+++ b/include/linux/mlx5/cq.h
@@ -185,7 +185,7 @@ static inline void mlx5_cq_put(struct mlx5_core_cq *cq)
}
int mlx5_core_create_cq(struct mlx5_core_dev *dev, struct mlx5_core_cq *cq,
- u32 *in, int inlen);
+ u32 *in, int inlen, u32 *out, int outlen);
int mlx5_core_destroy_cq(struct mlx5_core_dev *dev, struct mlx5_core_cq *cq);
int mlx5_core_query_cq(struct mlx5_core_dev *dev, struct mlx5_core_cq *cq,
u32 *out, int outlen);
--
2.20.1
^ permalink raw reply related
* [PATCH mlx5-next v2 05/13] net/mlx5: Report a CQ error event only when a handler was set
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Report a CQ error event only when a handler was set.
This enables mlx5_ib to not set a handler upon CQ creation and use some
other mechanism to get this event as of other events by the
mlx5_eq_notifier_register API.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eq.c b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
index c634a78d5cdd..678454535460 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
@@ -503,7 +503,8 @@ static int cq_err_event_notifier(struct notifier_block *nb,
return NOTIFY_OK;
}
- cq->event(cq, type);
+ if (cq->event)
+ cq->event(cq, type);
mlx5_cq_put(cq);
--
2.20.1
^ permalink raw reply related
* [PATCH mlx5-next v2 06/13] net/mlx5: Report EQE data upon CQ completion
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Report EQE data upon CQ completion to let upper layers use this data.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/cq.c | 2 +-
drivers/infiniband/hw/mlx5/main.c | 2 +-
drivers/infiniband/hw/mlx5/qp.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/cq.c | 5 +++--
drivers/net/ethernet/mellanox/mlx5/core/en.h | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c | 3 ++-
include/linux/mlx5/cq.h | 4 ++--
9 files changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index 857a554e2c29..6ff984db9e24 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -37,7 +37,7 @@
#include "mlx5_ib.h"
#include "srq.h"
-static void mlx5_ib_cq_comp(struct mlx5_core_cq *cq)
+static void mlx5_ib_cq_comp(struct mlx5_core_cq *cq, struct mlx5_eqe *eqe)
{
struct ib_cq *ibcq = &to_mibcq(cq)->ibcq;
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 6686f8f876f0..23997f744586 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -4463,7 +4463,7 @@ static void mlx5_ib_handle_internal_error(struct mlx5_ib_dev *ibdev)
* lock/unlock above locks Now need to arm all involved CQs.
*/
list_for_each_entry(mcq, &cq_armed_list, reset_notify) {
- mcq->comp(mcq);
+ mcq->comp(mcq, NULL);
}
spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
}
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 42375cdafd53..2a97619ed603 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -6406,7 +6406,7 @@ static void handle_drain_completion(struct ib_cq *cq,
/* Run the CQ handler - this makes sure that the drain WR will
* be processed if wasn't processed yet.
*/
- mcq->mcq.comp(&mcq->mcq);
+ mcq->mcq.comp(&mcq->mcq, NULL);
}
wait_for_completion(&sdrain->done);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/cq.c b/drivers/net/ethernet/mellanox/mlx5/core/cq.c
index 1bd4336392a2..818edc63e428 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/cq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/cq.c
@@ -58,7 +58,7 @@ void mlx5_cq_tasklet_cb(unsigned long data)
list_for_each_entry_safe(mcq, temp, &ctx->process_list,
tasklet_ctx.list) {
list_del_init(&mcq->tasklet_ctx.list);
- mcq->tasklet_ctx.comp(mcq);
+ mcq->tasklet_ctx.comp(mcq, NULL);
mlx5_cq_put(mcq);
if (time_after(jiffies, end))
break;
@@ -68,7 +68,8 @@ void mlx5_cq_tasklet_cb(unsigned long data)
tasklet_schedule(&ctx->task);
}
-static void mlx5_add_cq_to_tasklet(struct mlx5_core_cq *cq)
+static void mlx5_add_cq_to_tasklet(struct mlx5_core_cq *cq,
+ struct mlx5_eqe *eqe)
{
unsigned long flags;
struct mlx5_eq_tasklet *tasklet_ctx = cq->tasklet_ctx.priv;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index c6a8b615b56d..e7b337998547 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -781,7 +781,7 @@ netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
struct mlx5e_tx_wqe *wqe, u16 pi, bool xmit_more);
void mlx5e_trigger_irq(struct mlx5e_icosq *sq);
-void mlx5e_completion_event(struct mlx5_core_cq *mcq);
+void mlx5e_completion_event(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe);
void mlx5e_cq_error_event(struct mlx5_core_cq *mcq, enum mlx5_event event);
int mlx5e_napi_poll(struct napi_struct *napi, int budget);
bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index e6c434efbd46..1c021a299baa 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -134,7 +134,7 @@ int mlx5e_napi_poll(struct napi_struct *napi, int budget)
return work_done;
}
-void mlx5e_completion_event(struct mlx5_core_cq *mcq)
+void mlx5e_completion_event(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe)
{
struct mlx5e_cq *cq = container_of(mcq, struct mlx5e_cq, mcq);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eq.c b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
index 678454535460..41f25ea2e8d9 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
@@ -153,7 +153,7 @@ static int mlx5_eq_comp_int(struct notifier_block *nb,
cq = mlx5_eq_cq_get(eq, cqn);
if (likely(cq)) {
++cq->arm_sn;
- cq->comp(cq);
+ cq->comp(cq, eqe);
mlx5_cq_put(cq);
} else {
mlx5_core_warn(eq->dev, "Completion event for bogus CQ 0x%x\n", cqn);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
index dc7b9d9f274d..028891823f32 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
@@ -414,7 +414,8 @@ static void mlx5_fpga_conn_cq_tasklet(unsigned long data)
mlx5_fpga_conn_cqes(conn, MLX5_FPGA_CQ_BUDGET);
}
-static void mlx5_fpga_conn_cq_complete(struct mlx5_core_cq *mcq)
+static void mlx5_fpga_conn_cq_complete(struct mlx5_core_cq *mcq,
+ struct mlx5_eqe *eqe)
{
struct mlx5_fpga_conn *conn;
diff --git a/include/linux/mlx5/cq.h b/include/linux/mlx5/cq.h
index e44157a2b7db..40748fc1b11b 100644
--- a/include/linux/mlx5/cq.h
+++ b/include/linux/mlx5/cq.h
@@ -47,7 +47,7 @@ struct mlx5_core_cq {
struct completion free;
unsigned vector;
unsigned int irqn;
- void (*comp) (struct mlx5_core_cq *);
+ void (*comp)(struct mlx5_core_cq *cq, struct mlx5_eqe *eqe);
void (*event) (struct mlx5_core_cq *, enum mlx5_event);
u32 cons_index;
unsigned arm_sn;
@@ -55,7 +55,7 @@ struct mlx5_core_cq {
int pid;
struct {
struct list_head list;
- void (*comp)(struct mlx5_core_cq *);
+ void (*comp)(struct mlx5_core_cq *cq, struct mlx5_eqe *eqe);
void *priv;
} tasklet_ctx;
int reset_notify_added;
--
2.20.1
^ permalink raw reply related
* [PATCH mlx5-next v2 07/13] net/mlx5: Expose device definitions for object events
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Expose an extra device definitions for objects events.
It includes: object_type values for legacy objects and generic data
header for any other object.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
include/linux/mlx5/mlx5_ifc.h | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index 3ef716c054c2..5d2bf91e130b 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -91,6 +91,20 @@ enum {
enum {
MLX5_OBJ_TYPE_GENEVE_TLV_OPT = 0x000b,
+ MLX5_OBJ_TYPE_MKEY = 0xff01,
+ MLX5_OBJ_TYPE_QP = 0xff02,
+ MLX5_OBJ_TYPE_PSV = 0xff03,
+ MLX5_OBJ_TYPE_RMP = 0xff04,
+ MLX5_OBJ_TYPE_XRC_SRQ = 0xff05,
+ MLX5_OBJ_TYPE_RQ = 0xff06,
+ MLX5_OBJ_TYPE_SQ = 0xff07,
+ MLX5_OBJ_TYPE_TIR = 0xff08,
+ MLX5_OBJ_TYPE_TIS = 0xff09,
+ MLX5_OBJ_TYPE_DCT = 0xff0a,
+ MLX5_OBJ_TYPE_XRQ = 0xff0b,
+ MLX5_OBJ_TYPE_RQT = 0xff0e,
+ MLX5_OBJ_TYPE_FLOW_COUNTER = 0xff0f,
+ MLX5_OBJ_TYPE_CQ = 0xff10,
};
enum {
@@ -9755,4 +9769,11 @@ struct mlx5_ifc_query_esw_functions_out_bits {
u8 reserved_at_280[0x180];
};
+struct mlx5_ifc_affiliated_event_header_bits {
+ u8 reserved_at_0[0x10];
+ u8 obj_type[0x10];
+
+ u8 obj_id[0x20];
+};
+
#endif /* MLX5_IFC_H */
--
2.20.1
^ permalink raw reply related
* [PATCH rdma-next v2 08/13] IB/mlx5: Introduce MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Introduce MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD and its initial
implementation.
This object is from type class FD and will be used to read DEVX
async events.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/devx.c | 95 +++++++++++++++++++++++
include/uapi/rdma/mlx5_user_ioctl_cmds.h | 10 +++
include/uapi/rdma/mlx5_user_ioctl_verbs.h | 4 +
3 files changed, 109 insertions(+)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index 80b42d069328..89947b4f4f6f 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -33,6 +33,17 @@ struct devx_async_data {
struct mlx5_ib_uapi_devx_async_cmd_hdr hdr;
};
+struct devx_async_event_file {
+ struct ib_uobject uobj;
+ /* Head of events that are subscribed to this FD */
+ struct list_head subscribed_events_list;
+ spinlock_t lock;
+ wait_queue_head_t poll_wait;
+ struct list_head event_list;
+ struct mlx5_ib_dev *dev;
+ u8 omit_data:1;
+};
+
#define MLX5_MAX_DESTROY_INBOX_SIZE_DW MLX5_ST_SZ_DW(delete_fte_in)
struct devx_obj {
struct mlx5_core_dev *mdev;
@@ -1375,6 +1386,37 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC)(
return 0;
}
+static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC)(
+ struct uverbs_attr_bundle *attrs)
+{
+ struct ib_uobject *uobj = uverbs_attr_get_uobject(
+ attrs, MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_HANDLE);
+ struct devx_async_event_file *ev_file;
+ struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
+ &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
+ struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device);
+ u32 flags;
+ int err;
+
+ err = uverbs_get_flags32(&flags, attrs,
+ MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_FLAGS,
+ MLX5_IB_UAPI_DEVX_CR_EV_CH_FLAGS_OMIT_DATA);
+
+ if (err)
+ return err;
+
+ ev_file = container_of(uobj, struct devx_async_event_file,
+ uobj);
+ spin_lock_init(&ev_file->lock);
+ INIT_LIST_HEAD(&ev_file->event_list);
+ init_waitqueue_head(&ev_file->poll_wait);
+ if (flags & MLX5_IB_UAPI_DEVX_CR_EV_CH_FLAGS_OMIT_DATA)
+ ev_file->omit_data = 1;
+ INIT_LIST_HEAD(&ev_file->subscribed_events_list);
+ ev_file->dev = dev;
+ return 0;
+}
+
static void devx_query_callback(int status, struct mlx5_async_work *context)
{
struct devx_async_data *async_data =
@@ -1729,6 +1771,32 @@ static const struct file_operations devx_async_cmd_event_fops = {
.llseek = no_llseek,
};
+static ssize_t devx_async_event_read(struct file *filp, char __user *buf,
+ size_t count, loff_t *pos)
+{
+ return -EINVAL;
+}
+
+static __poll_t devx_async_event_poll(struct file *filp,
+ struct poll_table_struct *wait)
+{
+ return 0;
+}
+
+static int devx_async_event_close(struct inode *inode, struct file *filp)
+{
+ uverbs_close_fd(filp);
+ return 0;
+}
+
+static const struct file_operations devx_async_event_fops = {
+ .owner = THIS_MODULE,
+ .read = devx_async_event_read,
+ .poll = devx_async_event_poll,
+ .release = devx_async_event_close,
+ .llseek = no_llseek,
+};
+
static int devx_hot_unplug_async_cmd_event_file(struct ib_uobject *uobj,
enum rdma_remove_reason why)
{
@@ -1748,6 +1816,12 @@ static int devx_hot_unplug_async_cmd_event_file(struct ib_uobject *uobj,
return 0;
};
+static int devx_hot_unplug_async_event_file(struct ib_uobject *uobj,
+ enum rdma_remove_reason why)
+{
+ return 0;
+};
+
DECLARE_UVERBS_NAMED_METHOD(
MLX5_IB_METHOD_DEVX_UMEM_REG,
UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_UMEM_REG_HANDLE,
@@ -1913,6 +1987,24 @@ DECLARE_UVERBS_NAMED_OBJECT(
O_RDONLY),
&UVERBS_METHOD(MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC));
+DECLARE_UVERBS_NAMED_METHOD(
+ MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC,
+ UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_HANDLE,
+ MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
+ UVERBS_ACCESS_NEW,
+ UA_MANDATORY),
+ UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_FLAGS,
+ enum mlx5_ib_uapi_devx_create_event_channel_flags,
+ UA_MANDATORY));
+
+DECLARE_UVERBS_NAMED_OBJECT(
+ MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
+ UVERBS_TYPE_ALLOC_FD(sizeof(struct devx_async_event_file),
+ devx_hot_unplug_async_event_file,
+ &devx_async_event_fops, "[devx_async_event]",
+ O_RDONLY),
+ &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC));
+
static bool devx_is_supported(struct ib_device *device)
{
struct mlx5_ib_dev *dev = to_mdev(device);
@@ -1933,5 +2025,8 @@ const struct uapi_definition mlx5_ib_devx_defs[] = {
UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
+ UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
+ MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
+ UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
{},
};
diff --git a/include/uapi/rdma/mlx5_user_ioctl_cmds.h b/include/uapi/rdma/mlx5_user_ioctl_cmds.h
index d404c951954c..6ad8f4f11ddd 100644
--- a/include/uapi/rdma/mlx5_user_ioctl_cmds.h
+++ b/include/uapi/rdma/mlx5_user_ioctl_cmds.h
@@ -127,16 +127,26 @@ enum mlx5_ib_devx_async_cmd_fd_alloc_attrs {
MLX5_IB_ATTR_DEVX_ASYNC_CMD_FD_ALLOC_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
};
+enum mlx5_ib_devx_async_event_fd_alloc_attrs {
+ MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+ MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_FLAGS,
+};
+
enum mlx5_ib_devx_async_cmd_fd_methods {
MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC = (1U << UVERBS_ID_NS_SHIFT),
};
+enum mlx5_ib_devx_async_event_fd_methods {
+ MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC = (1U << UVERBS_ID_NS_SHIFT),
+};
+
enum mlx5_ib_objects {
MLX5_IB_OBJECT_DEVX = (1U << UVERBS_ID_NS_SHIFT),
MLX5_IB_OBJECT_DEVX_OBJ,
MLX5_IB_OBJECT_DEVX_UMEM,
MLX5_IB_OBJECT_FLOW_MATCHER,
MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
+ MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
};
enum mlx5_ib_flow_matcher_create_attrs {
diff --git a/include/uapi/rdma/mlx5_user_ioctl_verbs.h b/include/uapi/rdma/mlx5_user_ioctl_verbs.h
index a8f34c237458..b44691315d39 100644
--- a/include/uapi/rdma/mlx5_user_ioctl_verbs.h
+++ b/include/uapi/rdma/mlx5_user_ioctl_verbs.h
@@ -63,5 +63,9 @@ enum mlx5_ib_uapi_dm_type {
MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM,
};
+enum mlx5_ib_uapi_devx_create_event_channel_flags {
+ MLX5_IB_UAPI_DEVX_CR_EV_CH_FLAGS_OMIT_DATA = 1 << 0,
+};
+
#endif
--
2.20.1
^ permalink raw reply related
* [PATCH mlx5-next v2 02/13] net/mlx5: Use event mask based on device capabilities
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Use the reported device capabilities for the supported user events (i.e.
affiliated and un-affiliated) to set the EQ mask.
As the event mask can be up to 256 defined by 4 entries of u64 change
the applicable code to work accordingly.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/odp.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 40 ++++++++++++++++----
drivers/net/ethernet/mellanox/mlx5/core/fw.c | 6 +++
include/linux/mlx5/device.h | 6 ++-
include/linux/mlx5/eq.h | 2 +-
include/linux/mlx5/mlx5_ifc.h | 13 +++++--
6 files changed, 55 insertions(+), 14 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 600fe23e2eae..346a7ffd63aa 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -1559,9 +1559,9 @@ mlx5_ib_create_pf_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
eq->irq_nb.notifier_call = mlx5_ib_eq_pf_int;
param = (struct mlx5_eq_param) {
.irq_index = 0,
- .mask = 1 << MLX5_EVENT_TYPE_PAGE_FAULT,
.nent = MLX5_IB_NUM_PF_EQE,
};
+ param.mask[0] = 1ull << MLX5_EVENT_TYPE_PAGE_FAULT;
eq->core = mlx5_eq_create_generic(dev->mdev, ¶m);
if (IS_ERR(eq->core)) {
err = PTR_ERR(eq->core);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eq.c b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
index 8000d2a4a7e2..33f78d4d3724 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
@@ -256,6 +256,7 @@ create_map_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq,
int inlen;
u32 *in;
int err;
+ int i;
/* Init CQ table */
memset(cq_table, 0, sizeof(*cq_table));
@@ -283,10 +284,12 @@ create_map_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq,
mlx5_fill_page_array(&eq->buf, pas);
MLX5_SET(create_eq_in, in, opcode, MLX5_CMD_OP_CREATE_EQ);
- if (!param->mask && MLX5_CAP_GEN(dev, log_max_uctx))
+ if (!param->mask[0] && MLX5_CAP_GEN(dev, log_max_uctx))
MLX5_SET(create_eq_in, in, uid, MLX5_SHARED_RESOURCE_UID);
- MLX5_SET64(create_eq_in, in, event_bitmask, param->mask);
+ for (i = 0; i < 4; i++)
+ MLX5_ARRAY_SET64(create_eq_in, in, event_bitmask, i,
+ param->mask[i]);
eqc = MLX5_ADDR_OF(create_eq_in, in, eq_context_entry);
MLX5_SET(eqc, eqc, log_eq_size, ilog2(eq->nent));
@@ -507,7 +510,23 @@ static int cq_err_event_notifier(struct notifier_block *nb,
return NOTIFY_OK;
}
-static u64 gather_async_events_mask(struct mlx5_core_dev *dev)
+static void gather_user_async_events(struct mlx5_core_dev *dev, u64 mask[4])
+{
+ __be64 *user_unaffiliated_events;
+ __be64 *user_affiliated_events;
+ int i;
+
+ user_affiliated_events =
+ MLX5_CAP_DEV_EVENT(dev, user_affiliated_events);
+ user_unaffiliated_events =
+ MLX5_CAP_DEV_EVENT(dev, user_unaffiliated_events);
+
+ for (i = 0; i < 4; i++)
+ mask[i] |= be64_to_cpu(user_affiliated_events[i] |
+ user_unaffiliated_events[i]);
+}
+
+static void gather_async_events_mask(struct mlx5_core_dev *dev, u64 mask[4])
{
u64 async_event_mask = MLX5_ASYNC_EVENT_MASK;
@@ -544,7 +563,10 @@ static u64 gather_async_events_mask(struct mlx5_core_dev *dev)
async_event_mask |=
(1ull << MLX5_EVENT_TYPE_ESW_FUNCTIONS_CHANGED);
- return async_event_mask;
+ mask[0] = async_event_mask;
+
+ if (MLX5_CAP_GEN(dev, event_cap))
+ gather_user_async_events(dev, mask);
}
static int create_async_eqs(struct mlx5_core_dev *dev)
@@ -559,9 +581,10 @@ static int create_async_eqs(struct mlx5_core_dev *dev)
table->cmd_eq.irq_nb.notifier_call = mlx5_eq_async_int;
param = (struct mlx5_eq_param) {
.irq_index = 0,
- .mask = 1ull << MLX5_EVENT_TYPE_CMD,
.nent = MLX5_NUM_CMD_EQE,
};
+
+ param.mask[0] = 1ull << MLX5_EVENT_TYPE_CMD;
err = create_async_eq(dev, &table->cmd_eq.core, ¶m);
if (err) {
mlx5_core_warn(dev, "failed to create cmd EQ %d\n", err);
@@ -577,9 +600,10 @@ static int create_async_eqs(struct mlx5_core_dev *dev)
table->async_eq.irq_nb.notifier_call = mlx5_eq_async_int;
param = (struct mlx5_eq_param) {
.irq_index = 0,
- .mask = gather_async_events_mask(dev),
.nent = MLX5_NUM_ASYNC_EQE,
};
+
+ gather_async_events_mask(dev, param.mask);
err = create_async_eq(dev, &table->async_eq.core, ¶m);
if (err) {
mlx5_core_warn(dev, "failed to create async EQ %d\n", err);
@@ -595,9 +619,10 @@ static int create_async_eqs(struct mlx5_core_dev *dev)
table->pages_eq.irq_nb.notifier_call = mlx5_eq_async_int;
param = (struct mlx5_eq_param) {
.irq_index = 0,
- .mask = 1 << MLX5_EVENT_TYPE_PAGE_REQUEST,
.nent = /* TODO: sriov max_vf + */ 1,
};
+
+ param.mask[0] = 1ull << MLX5_EVENT_TYPE_PAGE_REQUEST;
err = create_async_eq(dev, &table->pages_eq.core, ¶m);
if (err) {
mlx5_core_warn(dev, "failed to create pages EQ %d\n", err);
@@ -789,7 +814,6 @@ static int create_comp_eqs(struct mlx5_core_dev *dev)
eq->irq_nb.notifier_call = mlx5_eq_comp_int;
param = (struct mlx5_eq_param) {
.irq_index = vecidx,
- .mask = 0,
.nent = nent,
};
err = create_map_eq(dev, &eq->core, ¶m);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fw.c b/drivers/net/ethernet/mellanox/mlx5/core/fw.c
index 1ab6f7e3bec6..05367f15c3a7 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fw.c
@@ -202,6 +202,12 @@ int mlx5_query_hca_caps(struct mlx5_core_dev *dev)
return err;
}
+ if (MLX5_CAP_GEN(dev, event_cap)) {
+ err = mlx5_core_get_caps(dev, MLX5_CAP_DEV_EVENT);
+ if (err)
+ return err;
+ }
+
return 0;
}
diff --git a/include/linux/mlx5/device.h b/include/linux/mlx5/device.h
index 5e760067ac41..0d1abe097627 100644
--- a/include/linux/mlx5/device.h
+++ b/include/linux/mlx5/device.h
@@ -351,7 +351,7 @@ enum mlx5_event {
MLX5_EVENT_TYPE_DEVICE_TRACER = 0x26,
- MLX5_EVENT_TYPE_MAX = MLX5_EVENT_TYPE_DEVICE_TRACER + 1,
+ MLX5_EVENT_TYPE_MAX = 0x100,
};
enum {
@@ -1077,6 +1077,7 @@ enum mlx5_cap_type {
MLX5_CAP_DEBUG,
MLX5_CAP_RESERVED_14,
MLX5_CAP_DEV_MEM,
+ MLX5_CAP_DEV_EVENT = 0x14,
/* NUM OF CAP Types */
MLX5_CAP_NUM
};
@@ -1255,6 +1256,9 @@ enum mlx5_qcam_feature_groups {
#define MLX5_CAP64_DEV_MEM(mdev, cap)\
MLX5_GET64(device_mem_cap, mdev->caps.hca_cur[MLX5_CAP_DEV_MEM], cap)
+#define MLX5_CAP_DEV_EVENT(mdev, cap)\
+ MLX5_ADDR_OF(device_event_cap, (mdev)->caps.hca_cur[MLX5_CAP_DEV_EVENT], cap)
+
enum {
MLX5_CMD_STAT_OK = 0x0,
MLX5_CMD_STAT_INT_ERR = 0x1,
diff --git a/include/linux/mlx5/eq.h b/include/linux/mlx5/eq.h
index 70e16dcfb4c4..e49d8c0d4f26 100644
--- a/include/linux/mlx5/eq.h
+++ b/include/linux/mlx5/eq.h
@@ -15,7 +15,7 @@ struct mlx5_core_dev;
struct mlx5_eq_param {
u8 irq_index;
int nent;
- u64 mask;
+ u64 mask[4];
};
struct mlx5_eq *
diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index 16348528fef6..3ef716c054c2 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -823,6 +823,12 @@ struct mlx5_ifc_device_mem_cap_bits {
u8 reserved_at_180[0x680];
};
+struct mlx5_ifc_device_event_cap_bits {
+ u8 user_affiliated_events[4][0x40];
+
+ u8 user_unaffiliated_events[4][0x40];
+};
+
enum {
MLX5_ATOMIC_CAPS_ATOMIC_SIZE_QP_1_BYTE = 0x0,
MLX5_ATOMIC_CAPS_ATOMIC_SIZE_QP_2_BYTES = 0x2,
@@ -980,7 +986,8 @@ struct mlx5_ifc_cmd_hca_cap_bits {
u8 log_max_srq_sz[0x8];
u8 log_max_qp_sz[0x8];
- u8 reserved_at_90[0x8];
+ u8 event_cap[0x1];
+ u8 reserved_at_91[0x7];
u8 prio_tag_required[0x1];
u8 reserved_at_99[0x2];
u8 log_max_qp[0x5];
@@ -7364,9 +7371,9 @@ struct mlx5_ifc_create_eq_in_bits {
u8 reserved_at_280[0x40];
- u8 event_bitmask[0x40];
+ u8 event_bitmask[4][0x40];
- u8 reserved_at_300[0x580];
+ u8 reserved_at_3c0[0x4c0];
u8 pas[0][0x40];
};
--
2.20.1
^ permalink raw reply related
* [PATCH rdma-next v2 10/13] IB/mlx5: Enable subscription for device events over DEVX
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Enable subscription for device events over DEVX.
Each subscription is added to the two level XA data structure according
to its event number and the DEVX object information in case was given
with the given target fd.
Those events will be reported over the given fd once will occur.
Downstream patches will mange the dispatching to any subscription.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/devx.c | 560 ++++++++++++++++++++++-
include/uapi/rdma/mlx5_user_ioctl_cmds.h | 9 +
2 files changed, 562 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index b7b17e882d82..6fb58339b291 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -14,6 +14,7 @@
#include <linux/mlx5/driver.h>
#include <linux/mlx5/fs.h>
#include "mlx5_ib.h"
+#include <linux/xarray.h>
#define UVERBS_MODULE_NAME mlx5_ib
#include <rdma/uverbs_named_ioctl.h>
@@ -33,6 +34,40 @@ struct devx_async_data {
struct mlx5_ib_uapi_devx_async_cmd_hdr hdr;
};
+/* first level XA value data structure */
+struct devx_event {
+ struct xarray object_ids; /* second XA level, Key = object id */
+ struct list_head unaffiliated_list;
+};
+
+/* second level XA value data structure */
+struct devx_obj_event {
+ struct rcu_head rcu;
+ struct list_head obj_sub_list;
+};
+
+struct devx_event_subscription {
+ struct list_head file_list; /* headed in ev_file->
+ * subscribed_events_list
+ */
+ struct list_head xa_list; /* headed in devx_event->unaffiliated_list or
+ * devx_obj_event->obj_sub_list
+ */
+ struct list_head obj_list; /* headed in devx_object */
+ struct list_head event_list; /* headed in ev_file->event_list or in
+ * temp list via subscription
+ */
+
+ u8 is_cleaned:1;
+ u32 xa_key_level1;
+ u32 xa_key_level2;
+ struct rcu_head rcu;
+ u64 cookie;
+ struct devx_async_event_file *ev_file;
+ struct file *filp; /* Upon hot unplug we need a direct access to */
+ struct eventfd_ctx *eventfd;
+};
+
struct devx_async_event_file {
struct ib_uobject uobj;
/* Head of events that are subscribed to this FD */
@@ -55,6 +90,7 @@ struct devx_obj {
struct mlx5_ib_devx_mr devx_mr;
struct mlx5_core_dct core_dct;
};
+ struct list_head event_sub; /* holds devx_event_subscription entries */
};
struct devx_umem {
@@ -160,6 +196,104 @@ bool mlx5_ib_devx_is_flow_counter(void *obj, u32 *counter_id)
return false;
}
+static bool is_legacy_unaffiliated_event_num(u16 event_num)
+{
+ switch (event_num) {
+ case MLX5_EVENT_TYPE_PORT_CHANGE:
+ return true;
+ default:
+ return false;
+ }
+}
+
+static bool is_legacy_obj_event_num(u16 event_num)
+{
+ switch (event_num) {
+ case MLX5_EVENT_TYPE_PATH_MIG:
+ case MLX5_EVENT_TYPE_COMM_EST:
+ case MLX5_EVENT_TYPE_SQ_DRAINED:
+ case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
+ case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
+ case MLX5_EVENT_TYPE_CQ_ERROR:
+ case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
+ case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
+ case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+ case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
+ case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
+ case MLX5_EVENT_TYPE_DCT_DRAINED:
+ case MLX5_EVENT_TYPE_COMP:
+ return true;
+ default:
+ return false;
+ }
+}
+
+static u16 get_legacy_obj_type(u16 opcode)
+{
+ switch (opcode) {
+ case MLX5_CMD_OP_CREATE_RQ:
+ return MLX5_EVENT_QUEUE_TYPE_RQ;
+ case MLX5_CMD_OP_CREATE_QP:
+ return MLX5_EVENT_QUEUE_TYPE_QP;
+ case MLX5_CMD_OP_CREATE_SQ:
+ return MLX5_EVENT_QUEUE_TYPE_SQ;
+ case MLX5_CMD_OP_CREATE_DCT:
+ return MLX5_EVENT_QUEUE_TYPE_DCT;
+ default:
+ return 0;
+ }
+}
+
+static u16 get_dec_obj_type(struct devx_obj *obj, u16 event_num)
+{
+ u16 opcode;
+
+ opcode = (obj->obj_id >> 32) & 0xffff;
+
+ if (is_legacy_obj_event_num(event_num))
+ return get_legacy_obj_type(opcode);
+
+ switch (opcode) {
+ case MLX5_CMD_OP_CREATE_GENERAL_OBJECT:
+ return (obj->obj_id >> 48);
+ case MLX5_CMD_OP_CREATE_RQ:
+ return MLX5_OBJ_TYPE_RQ;
+ case MLX5_CMD_OP_CREATE_QP:
+ return MLX5_OBJ_TYPE_QP;
+ case MLX5_CMD_OP_CREATE_SQ:
+ return MLX5_OBJ_TYPE_SQ;
+ case MLX5_CMD_OP_CREATE_DCT:
+ return MLX5_OBJ_TYPE_DCT;
+ case MLX5_CMD_OP_CREATE_TIR:
+ return MLX5_OBJ_TYPE_TIR;
+ case MLX5_CMD_OP_CREATE_TIS:
+ return MLX5_OBJ_TYPE_TIS;
+ case MLX5_CMD_OP_CREATE_PSV:
+ return MLX5_OBJ_TYPE_PSV;
+ case MLX5_OBJ_TYPE_MKEY:
+ return MLX5_OBJ_TYPE_MKEY;
+ case MLX5_CMD_OP_CREATE_RMP:
+ return MLX5_OBJ_TYPE_RMP;
+ case MLX5_CMD_OP_CREATE_XRC_SRQ:
+ return MLX5_OBJ_TYPE_XRC_SRQ;
+ case MLX5_CMD_OP_CREATE_XRQ:
+ return MLX5_OBJ_TYPE_XRQ;
+ case MLX5_CMD_OP_CREATE_RQT:
+ return MLX5_OBJ_TYPE_RQT;
+ case MLX5_CMD_OP_ALLOC_FLOW_COUNTER:
+ return MLX5_OBJ_TYPE_FLOW_COUNTER;
+ case MLX5_CMD_OP_CREATE_CQ:
+ return MLX5_OBJ_TYPE_CQ;
+ default:
+ return 0;
+ }
+}
+
+static u32 get_dec_obj_id(u64 obj_id)
+{
+ return (obj_id & 0xffffffff);
+}
+
/*
* As the obj_id in the firmware is not globally unique the object type
* must be considered upon checking for a valid object id.
@@ -1136,14 +1270,47 @@ static void devx_cleanup_mkey(struct devx_obj *obj)
write_unlock_irqrestore(&table->lock, flags);
}
+static void devx_cleanup_subscription(struct mlx5_ib_dev *dev,
+ struct devx_event_subscription *sub)
+{
+ struct devx_event *event;
+ struct devx_obj_event *xa_val_level2;
+
+ if (sub->is_cleaned)
+ return;
+
+ sub->is_cleaned = 1;
+ list_del_rcu(&sub->xa_list);
+
+ if (list_empty(&sub->obj_list))
+ return;
+
+ list_del_rcu(&sub->obj_list);
+ /* check whether key level 1 for this obj_sub_list is empty */
+ event = xa_load(&dev->devx_event_table.event_xa,
+ sub->xa_key_level1);
+ WARN_ON(!event);
+
+ xa_val_level2 = xa_load(&event->object_ids, sub->xa_key_level2);
+ if (list_empty(&xa_val_level2->obj_sub_list)) {
+ xa_erase(&event->object_ids,
+ sub->xa_key_level2);
+ kfree_rcu(xa_val_level2, rcu);
+ }
+}
+
static int devx_obj_cleanup(struct ib_uobject *uobject,
enum rdma_remove_reason why,
struct uverbs_attr_bundle *attrs)
{
u32 out[MLX5_ST_SZ_DW(general_obj_out_cmd_hdr)];
+ struct mlx5_devx_event_table *devx_event_table;
struct devx_obj *obj = uobject->object;
+ struct devx_event_subscription *sub_entry, *tmp;
+ struct mlx5_ib_dev *dev;
int ret;
+ dev = mlx5_udata_to_mdev(&attrs->driver_udata);
if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY)
devx_cleanup_mkey(obj);
@@ -1155,10 +1322,14 @@ static int devx_obj_cleanup(struct ib_uobject *uobject,
if (ib_is_destroy_retryable(ret, why, uobject))
return ret;
- if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) {
- struct mlx5_ib_dev *dev =
- mlx5_udata_to_mdev(&attrs->driver_udata);
+ devx_event_table = &dev->devx_event_table;
+
+ mutex_lock(&devx_event_table->event_xa_lock);
+ list_for_each_entry_safe(sub_entry, tmp, &obj->event_sub, obj_list)
+ devx_cleanup_subscription(dev, sub_entry);
+ mutex_unlock(&devx_event_table->event_xa_lock);
+ if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) {
call_srcu(&dev->mr_srcu, &obj->devx_mr.rcu,
devx_free_indirect_mkey);
return ret;
@@ -1230,6 +1401,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
uobj->object = obj;
obj->mdev = dev->mdev;
+ INIT_LIST_HEAD(&obj->event_sub);
devx_obj_build_destroy_cmd(cmd_in, cmd_out, obj->dinbox, &obj->dinlen,
&obj_id);
WARN_ON(obj->dinlen > MLX5_MAX_DESTROY_INBOX_SIZE_DW * sizeof(u32));
@@ -1414,6 +1586,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC)(
ev_file->omit_data = 1;
INIT_LIST_HEAD(&ev_file->subscribed_events_list);
ev_file->dev = dev;
+ get_device(&dev->ib_dev.dev);
return 0;
}
@@ -1526,6 +1699,331 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY)(
return err;
}
+static void
+subscribe_event_xa_dealloc(struct mlx5_devx_event_table *devx_event_table,
+ u32 key_level1,
+ bool is_level2,
+ u32 key_level2)
+{
+ struct devx_event *event;
+ struct devx_obj_event *xa_val_level2;
+
+ /* Level 1 is valid for future use, no need to free */
+ if (!is_level2)
+ return;
+
+ event = xa_load(&devx_event_table->event_xa, key_level1);
+ WARN_ON(!event);
+
+ xa_val_level2 = xa_load(&event->object_ids,
+ key_level2);
+ if (list_empty(&xa_val_level2->obj_sub_list)) {
+ xa_erase(&event->object_ids,
+ key_level2);
+ kfree_rcu(xa_val_level2, rcu);
+ }
+}
+
+static int
+subscribe_event_xa_alloc(struct mlx5_devx_event_table *devx_event_table,
+ u32 key_level1,
+ bool is_level2,
+ u32 key_level2)
+{
+ struct devx_obj_event *obj_event;
+ struct devx_event *event;
+ int err;
+
+ event = xa_load(&devx_event_table->event_xa, key_level1);
+ if (!event) {
+ event = kzalloc(sizeof(*event), GFP_KERNEL);
+ if (!event)
+ return -ENOMEM;
+
+ INIT_LIST_HEAD(&event->unaffiliated_list);
+ xa_init(&event->object_ids);
+
+ err = xa_insert(&devx_event_table->event_xa,
+ key_level1,
+ event,
+ GFP_KERNEL);
+ if (err) {
+ kfree(event);
+ return err;
+ }
+ }
+
+ if (!is_level2)
+ return 0;
+
+ obj_event = xa_load(&event->object_ids, key_level2);
+ if (!obj_event) {
+ obj_event = kzalloc(sizeof(*obj_event), GFP_KERNEL);
+ if (!obj_event)
+ /* Level1 is valid for future use, no need to free */
+ return -ENOMEM;
+
+ err = xa_insert(&event->object_ids,
+ key_level2,
+ obj_event,
+ GFP_KERNEL);
+ if (err)
+ return err;
+ INIT_LIST_HEAD(&obj_event->obj_sub_list);
+ }
+
+ return 0;
+}
+
+static bool is_valid_events_legacy(int num_events, u16 *event_type_num_list,
+ struct devx_obj *obj)
+{
+ int i;
+
+ for (i = 0; i < num_events; i++) {
+ if (obj) {
+ if (!is_legacy_obj_event_num(event_type_num_list[i]))
+ return false;
+ } else if (!is_legacy_unaffiliated_event_num(
+ event_type_num_list[i])) {
+ return false;
+ }
+ }
+
+ return true;
+}
+
+#define MAX_SUPP_EVENT_NUM 255
+static bool is_valid_events(struct mlx5_core_dev *dev,
+ int num_events, u16 *event_type_num_list,
+ struct devx_obj *obj)
+{
+ __be64 *aff_events;
+ __be64 *unaff_events;
+ int mask_entry;
+ int mask_bit;
+ int i;
+
+ if (MLX5_CAP_GEN(dev, event_cap)) {
+ aff_events = MLX5_CAP_DEV_EVENT(dev,
+ user_affiliated_events);
+ unaff_events = MLX5_CAP_DEV_EVENT(dev,
+ user_unaffiliated_events);
+ } else {
+ return is_valid_events_legacy(num_events, event_type_num_list,
+ obj);
+ }
+
+ for (i = 0; i < num_events; i++) {
+ if (event_type_num_list[i] > MAX_SUPP_EVENT_NUM)
+ return false;
+
+ mask_entry = event_type_num_list[i] / 64;
+ mask_bit = event_type_num_list[i] % 64;
+
+ if (obj) {
+ /* CQ completion */
+ if (event_type_num_list[i] == 0)
+ continue;
+
+ if (!(be64_to_cpu(aff_events[mask_entry]) &
+ (1ull << mask_bit)))
+ return false;
+
+ continue;
+ }
+
+ if (!(be64_to_cpu(unaff_events[mask_entry]) &
+ (1ull << mask_bit)))
+ return false;
+ }
+
+ return true;
+}
+
+#define MAX_NUM_EVENTS 16
+static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT)(
+ struct uverbs_attr_bundle *attrs)
+{
+ struct ib_uobject *devx_uobj = uverbs_attr_get_uobject(
+ attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_OBJ_HANDLE);
+ struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
+ &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
+ struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device);
+ struct ib_uobject *fd_uobj;
+ struct devx_obj *obj = NULL;
+ struct devx_async_event_file *ev_file;
+ struct mlx5_devx_event_table *devx_event_table = &dev->devx_event_table;
+ u16 *event_type_num_list;
+ struct devx_event_subscription *event_sub, *tmp_sub;
+ struct list_head sub_list;
+ int redirect_fd;
+ bool use_eventfd = false;
+ int num_events;
+ int num_alloc_xa_entries = 0;
+ u16 obj_type = 0;
+ u64 cookie = 0;
+ u32 obj_id = 0;
+ int err;
+ int i;
+
+ if (!c->devx_uid)
+ return -EINVAL;
+
+ if (!IS_ERR(devx_uobj)) {
+ obj = (struct devx_obj *)devx_uobj->object;
+ if (obj)
+ obj_id = get_dec_obj_id(obj->obj_id);
+ }
+
+ fd_uobj = uverbs_attr_get_uobject(attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_HANDLE);
+ if (IS_ERR(fd_uobj))
+ return PTR_ERR(fd_uobj);
+
+ ev_file = container_of(fd_uobj, struct devx_async_event_file,
+ uobj);
+
+ if (uverbs_attr_is_valid(attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM)) {
+ err = uverbs_copy_from(&redirect_fd, attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM);
+ if (err)
+ return err;
+
+ use_eventfd = true;
+ }
+
+ if (uverbs_attr_is_valid(attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE)) {
+ if (use_eventfd)
+ return -EINVAL;
+
+ err = uverbs_copy_from(&cookie, attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE);
+ if (err)
+ return err;
+ }
+
+ num_events = uverbs_attr_ptr_get_array_size(
+ attrs, MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST,
+ sizeof(u16));
+
+ if (num_events < 0)
+ return num_events;
+
+ if (num_events > MAX_NUM_EVENTS)
+ return -EINVAL;
+
+ event_type_num_list = uverbs_attr_get_alloced_ptr(attrs,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST);
+
+ if (!is_valid_events(dev->mdev, num_events, event_type_num_list, obj))
+ return -EINVAL;
+
+ INIT_LIST_HEAD(&sub_list);
+
+ /* Protect from concurrent subscriptions to same XA entries to allow
+ * both to succeed
+ */
+ mutex_lock(&devx_event_table->event_xa_lock);
+ for (i = 0; i < num_events; i++) {
+ u32 key_level1;
+
+ if (obj)
+ obj_type = get_dec_obj_type(obj,
+ event_type_num_list[i]);
+ key_level1 = event_type_num_list[i] | obj_type << 16;
+
+ err = subscribe_event_xa_alloc(devx_event_table,
+ key_level1,
+ obj,
+ obj_id);
+ if (err)
+ goto err;
+
+ num_alloc_xa_entries++;
+ event_sub = kzalloc(sizeof(*event_sub), GFP_KERNEL);
+ if (!event_sub)
+ goto err;
+
+ list_add_tail(&event_sub->event_list, &sub_list);
+ if (use_eventfd) {
+ event_sub->eventfd =
+ eventfd_ctx_fdget(redirect_fd);
+
+ if (IS_ERR(event_sub)) {
+ err = PTR_ERR(event_sub->eventfd);
+ event_sub->eventfd = NULL;
+ goto err;
+ }
+ }
+
+ event_sub->cookie = cookie;
+ event_sub->ev_file = ev_file;
+ event_sub->filp = fd_uobj->object;
+ /* May be needed upon cleanup the devx object/subscription */
+ event_sub->xa_key_level1 = key_level1;
+ event_sub->xa_key_level2 = obj_id;
+ INIT_LIST_HEAD(&event_sub->obj_list);
+ }
+
+ /* Once all the allocations and the XA data insertions were done we
+ * can go ahead and add all the subscriptions to the relevant lists
+ * without concern of a failure.
+ */
+ list_for_each_entry_safe(event_sub, tmp_sub, &sub_list, event_list) {
+ struct devx_event *event;
+ struct devx_obj_event *obj_event;
+
+ list_del_init(&event_sub->event_list);
+
+ spin_lock_irq(&ev_file->lock);
+ list_add_tail_rcu(&event_sub->file_list,
+ &ev_file->subscribed_events_list);
+ spin_unlock_irq(&ev_file->lock);
+
+ event = xa_load(&devx_event_table->event_xa,
+ event_sub->xa_key_level1);
+ WARN_ON(!event);
+
+ if (!obj) {
+ list_add_tail_rcu(&event_sub->xa_list,
+ &event->unaffiliated_list);
+ continue;
+ }
+
+ obj_event = xa_load(&event->object_ids, obj_id);
+ WARN_ON(!obj_event);
+ list_add_tail_rcu(&event_sub->xa_list,
+ &obj_event->obj_sub_list);
+ list_add_tail_rcu(&event_sub->obj_list,
+ &obj->event_sub);
+ }
+
+ mutex_unlock(&devx_event_table->event_xa_lock);
+ return 0;
+
+err:
+ list_for_each_entry_safe(event_sub, tmp_sub, &sub_list, event_list) {
+ list_del(&event_sub->event_list);
+
+ subscribe_event_xa_dealloc(devx_event_table,
+ event_sub->xa_key_level1,
+ obj,
+ obj_id);
+
+ if (event_sub->eventfd)
+ eventfd_ctx_put(event_sub->eventfd);
+
+ kfree(event_sub);
+ }
+
+ mutex_unlock(&devx_event_table->event_xa_lock);
+ return err;
+}
+
static int devx_umem_get(struct mlx5_ib_dev *dev, struct ib_ucontext *ucontext,
struct uverbs_attr_bundle *attrs,
struct devx_umem *obj)
@@ -1692,14 +2190,21 @@ void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev)
void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev)
{
struct mlx5_devx_event_table *table = &dev->devx_event_table;
+ struct devx_event_subscription *sub, *tmp;
+ struct devx_event *event;
void *entry;
unsigned long id;
mlx5_eq_notifier_unregister(dev->mdev, &table->devx_nb);
-
- xa_for_each(&table->event_xa, id, entry)
+ mutex_lock(&dev->devx_event_table.event_xa_lock);
+ xa_for_each(&table->event_xa, id, entry) {
+ event = entry;
+ list_for_each_entry_safe(sub, tmp, &event->unaffiliated_list,
+ xa_list)
+ devx_cleanup_subscription(dev, sub);
kfree(entry);
-
+ }
+ mutex_unlock(&dev->devx_event_table.event_xa_lock);
xa_destroy(&table->event_xa);
}
@@ -1815,7 +2320,26 @@ static __poll_t devx_async_event_poll(struct file *filp,
static int devx_async_event_close(struct inode *inode, struct file *filp)
{
+ struct devx_async_event_file *ev_file = filp->private_data;
+ struct devx_event_subscription *event_sub, *event_sub_tmp;
+
+ mutex_lock(&ev_file->dev->devx_event_table.event_xa_lock);
+ /* delete the subscriptions which are related to this FD */
+ list_for_each_entry_safe(event_sub, event_sub_tmp,
+ &ev_file->subscribed_events_list, file_list) {
+ devx_cleanup_subscription(ev_file->dev, event_sub);
+ if (event_sub->eventfd)
+ eventfd_ctx_put(event_sub->eventfd);
+
+ list_del_rcu(&event_sub->file_list);
+ /* subscription may not be used by the read API any more */
+ kfree_rcu(event_sub, rcu);
+ }
+
+ mutex_unlock(&ev_file->dev->devx_event_table.event_xa_lock);
+
uverbs_close_fd(filp);
+ put_device(&ev_file->dev->ib_dev.dev);
return 0;
}
@@ -1983,10 +2507,32 @@ DECLARE_UVERBS_NAMED_METHOD(
UVERBS_ATTR_TYPE(u64),
UA_MANDATORY));
+DECLARE_UVERBS_NAMED_METHOD(
+ MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT,
+ UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_HANDLE,
+ MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
+ UVERBS_ACCESS_READ,
+ UA_MANDATORY),
+ UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_OBJ_HANDLE,
+ MLX5_IB_OBJECT_DEVX_OBJ,
+ UVERBS_ACCESS_READ,
+ UA_OPTIONAL),
+ UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST,
+ UVERBS_ATTR_MIN_SIZE(sizeof(u16)),
+ UA_MANDATORY,
+ UA_ALLOC_AND_COPY),
+ UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE,
+ UVERBS_ATTR_TYPE(u64),
+ UA_OPTIONAL),
+ UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM,
+ UVERBS_ATTR_TYPE(u32),
+ UA_OPTIONAL));
+
DECLARE_UVERBS_GLOBAL_METHODS(MLX5_IB_OBJECT_DEVX,
&UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OTHER),
&UVERBS_METHOD(MLX5_IB_METHOD_DEVX_QUERY_UAR),
- &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_QUERY_EQN));
+ &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_QUERY_EQN),
+ &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT));
DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_DEVX_OBJ,
UVERBS_TYPE_ALLOC_IDR(devx_obj_cleanup),
diff --git a/include/uapi/rdma/mlx5_user_ioctl_cmds.h b/include/uapi/rdma/mlx5_user_ioctl_cmds.h
index 6ad8f4f11ddd..d0da070cf0ab 100644
--- a/include/uapi/rdma/mlx5_user_ioctl_cmds.h
+++ b/include/uapi/rdma/mlx5_user_ioctl_cmds.h
@@ -51,6 +51,7 @@ enum mlx5_ib_devx_methods {
MLX5_IB_METHOD_DEVX_OTHER = (1U << UVERBS_ID_NS_SHIFT),
MLX5_IB_METHOD_DEVX_QUERY_UAR,
MLX5_IB_METHOD_DEVX_QUERY_EQN,
+ MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT,
};
enum mlx5_ib_devx_other_attrs {
@@ -93,6 +94,14 @@ enum mlx5_ib_devx_obj_query_async_attrs {
MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_OUT_LEN,
};
+enum mlx5_ib_devx_subscribe_event_attrs {
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_OBJ_HANDLE,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM,
+ MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE,
+};
+
enum mlx5_ib_devx_query_eqn_attrs {
MLX5_IB_ATTR_DEVX_QUERY_EQN_USER_VEC = (1U << UVERBS_ID_NS_SHIFT),
MLX5_IB_ATTR_DEVX_QUERY_EQN_DEV_EQN,
--
2.20.1
^ permalink raw reply related
* [PATCH rdma-next v2 11/13] IB/mlx5: Implement DEVX dispatching event
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Implement DEVX dispatching event by looking up for the applicable
subscriptions for the reported event and using their target fd to
signal/set the event.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/devx.c | 303 +++++++++++++++++++++-
include/uapi/rdma/mlx5_user_ioctl_verbs.h | 5 +
2 files changed, 305 insertions(+), 3 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index 6fb58339b291..de068fe94363 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -34,6 +34,11 @@ struct devx_async_data {
struct mlx5_ib_uapi_devx_async_cmd_hdr hdr;
};
+struct devx_async_event_data {
+ struct list_head list; /* headed in ev_file->event_list */
+ struct mlx5_ib_uapi_devx_async_event_hdr hdr;
+};
+
/* first level XA value data structure */
struct devx_event {
struct xarray object_ids; /* second XA level, Key = object id */
@@ -77,6 +82,8 @@ struct devx_async_event_file {
struct list_head event_list;
struct mlx5_ib_dev *dev;
u8 omit_data:1;
+ u8 is_overflow_err:1;
+ u8 is_destroyed:1;
};
#define MLX5_MAX_DESTROY_INBOX_SIZE_DW MLX5_ST_SZ_DW(delete_fte_in)
@@ -289,6 +296,29 @@ static u16 get_dec_obj_type(struct devx_obj *obj, u16 event_num)
}
}
+static u16 get_event_obj_type(unsigned long event_type, struct mlx5_eqe *eqe)
+{
+ switch (event_type) {
+ case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
+ case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
+ case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+ case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
+ case MLX5_EVENT_TYPE_PATH_MIG:
+ case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
+ case MLX5_EVENT_TYPE_COMM_EST:
+ case MLX5_EVENT_TYPE_SQ_DRAINED:
+ case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
+ case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
+ return eqe->data.qp_srq.type;
+ case MLX5_EVENT_TYPE_CQ_ERROR:
+ return 0;
+ case MLX5_EVENT_TYPE_DCT_DRAINED:
+ return MLX5_EVENT_QUEUE_TYPE_DCT;
+ default:
+ return MLX5_GET(affiliated_event_header, &eqe->data, obj_type);
+ }
+}
+
static u32 get_dec_obj_id(u64 obj_id)
{
return (obj_id & 0xffffffff);
@@ -2171,10 +2201,170 @@ static int devx_umem_cleanup(struct ib_uobject *uobject,
return 0;
}
+static bool is_unaffiliated_event(struct mlx5_core_dev *dev,
+ unsigned long event_type)
+{
+ __be64 *unaff_events;
+ int mask_entry;
+ int mask_bit;
+
+ if (!MLX5_CAP_GEN(dev, event_cap))
+ return is_legacy_unaffiliated_event_num(event_type);
+
+ unaff_events = MLX5_CAP_DEV_EVENT(dev,
+ user_unaffiliated_events);
+ WARN_ON(event_type > MAX_SUPP_EVENT_NUM);
+
+ mask_entry = event_type / 64;
+ mask_bit = event_type % 64;
+
+ if (!(be64_to_cpu(unaff_events[mask_entry]) & (1ull << mask_bit)))
+ return false;
+
+ return true;
+}
+
+static u32 devx_get_obj_id_from_event(unsigned long event_type, void *data)
+{
+ struct mlx5_eqe *eqe = data;
+ u32 obj_id = 0;
+
+ switch (event_type) {
+ case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
+ case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
+ case MLX5_EVENT_TYPE_PATH_MIG:
+ case MLX5_EVENT_TYPE_COMM_EST:
+ case MLX5_EVENT_TYPE_SQ_DRAINED:
+ case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
+ case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
+ case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
+ case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+ case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
+ obj_id = be32_to_cpu(eqe->data.qp_srq.qp_srq_n) & 0xffffff;
+ break;
+ case MLX5_EVENT_TYPE_DCT_DRAINED:
+ obj_id = be32_to_cpu(eqe->data.dct.dctn) & 0xffffff;
+ break;
+ case MLX5_EVENT_TYPE_CQ_ERROR:
+ obj_id = be32_to_cpu(eqe->data.cq_err.cqn) & 0xffffff;
+ break;
+ default:
+ obj_id = MLX5_GET(affiliated_event_header, &eqe->data, obj_id);
+ break;
+ }
+
+ return obj_id;
+}
+
+static int deliver_event(struct devx_event_subscription *event_sub,
+ const void *data)
+{
+ struct devx_async_event_file *ev_file;
+ struct devx_async_event_data *event_data;
+ unsigned long flags;
+
+ ev_file = event_sub->ev_file;
+
+ if (ev_file->omit_data) {
+ spin_lock_irqsave(&ev_file->lock, flags);
+ if (!list_empty(&event_sub->event_list)) {
+ spin_unlock_irqrestore(&ev_file->lock, flags);
+ return 0;
+ }
+
+ list_add_tail(&event_sub->event_list, &ev_file->event_list);
+ spin_unlock_irqrestore(&ev_file->lock, flags);
+ wake_up_interruptible(&ev_file->poll_wait);
+ return 0;
+ }
+
+ event_data = kzalloc(sizeof(*event_data) + sizeof(struct mlx5_eqe),
+ GFP_ATOMIC);
+ if (!event_data) {
+ spin_lock_irqsave(&ev_file->lock, flags);
+ ev_file->is_overflow_err = 1;
+ spin_unlock_irqrestore(&ev_file->lock, flags);
+ return -ENOMEM;
+ }
+
+ event_data->hdr.cookie = event_sub->cookie;
+ memcpy(event_data->hdr.out_data, data, sizeof(struct mlx5_eqe));
+
+ spin_lock_irqsave(&ev_file->lock, flags);
+ list_add_tail(&event_data->list, &ev_file->event_list);
+ spin_unlock_irqrestore(&ev_file->lock, flags);
+ wake_up_interruptible(&ev_file->poll_wait);
+
+ return 0;
+}
+
+static void dispatch_event_fd(struct list_head *fd_list,
+ const void *data)
+{
+ struct devx_event_subscription *item;
+
+ list_for_each_entry_rcu(item, fd_list, xa_list) {
+ if (!get_file_rcu(item->filp))
+ continue;
+
+ if (item->eventfd) {
+ eventfd_signal(item->eventfd, 1);
+ fput(item->filp);
+ continue;
+ }
+
+ deliver_event(item, data);
+ fput(item->filp);
+ }
+}
+
static int devx_event_notifier(struct notifier_block *nb,
unsigned long event_type, void *data)
{
- return NOTIFY_DONE;
+ struct mlx5_devx_event_table *table;
+ struct mlx5_ib_dev *dev;
+ struct devx_event *event;
+ struct devx_obj_event *obj_event;
+ u16 obj_type = 0;
+ bool is_unaffiliated;
+ u32 obj_id;
+
+ /* Explicit filtering to kernel events which may occur frequently */
+ if (event_type == MLX5_EVENT_TYPE_CMD ||
+ event_type == MLX5_EVENT_TYPE_PAGE_REQUEST)
+ return NOTIFY_OK;
+
+ table = container_of(nb, struct mlx5_devx_event_table, devx_nb.nb);
+ dev = container_of(table, struct mlx5_ib_dev, devx_event_table);
+ is_unaffiliated = is_unaffiliated_event(dev->mdev, event_type);
+
+ if (!is_unaffiliated)
+ obj_type = get_event_obj_type(event_type, data);
+
+ rcu_read_lock();
+ event = xa_load(&table->event_xa, event_type | (obj_type << 16));
+ if (!event) {
+ rcu_read_unlock();
+ return NOTIFY_DONE;
+ }
+
+ if (is_unaffiliated) {
+ dispatch_event_fd(&event->unaffiliated_list, data);
+ rcu_read_unlock();
+ return NOTIFY_OK;
+ }
+
+ obj_id = devx_get_obj_id_from_event(event_type, data);
+ obj_event = xa_load(&event->object_ids, obj_id);
+ if (!obj_event) {
+ rcu_read_unlock();
+ return NOTIFY_DONE;
+ }
+
+ dispatch_event_fd(&obj_event->obj_sub_list, data);
+
+ rcu_read_unlock();
+ return NOTIFY_OK;
}
void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev)
@@ -2309,19 +2499,108 @@ static const struct file_operations devx_async_cmd_event_fops = {
static ssize_t devx_async_event_read(struct file *filp, char __user *buf,
size_t count, loff_t *pos)
{
- return -EINVAL;
+ struct devx_async_event_file *ev_file = filp->private_data;
+ struct devx_event_subscription *event_sub;
+ struct devx_async_event_data *uninitialized_var(event);
+ int ret = 0;
+ size_t eventsz;
+ bool omit_data;
+ void *event_data;
+
+ omit_data = ev_file->omit_data;
+
+ spin_lock_irq(&ev_file->lock);
+
+ if (ev_file->is_overflow_err) {
+ ev_file->is_overflow_err = 0;
+ spin_unlock_irq(&ev_file->lock);
+ return -EOVERFLOW;
+ }
+
+ if (ev_file->is_destroyed) {
+ spin_unlock_irq(&ev_file->lock);
+ return -EIO;
+ }
+
+ while (list_empty(&ev_file->event_list)) {
+ spin_unlock_irq(&ev_file->lock);
+
+ if (filp->f_flags & O_NONBLOCK)
+ return -EAGAIN;
+
+ if (wait_event_interruptible(ev_file->poll_wait,
+ (!list_empty(&ev_file->event_list) ||
+ ev_file->is_destroyed))) {
+ return -ERESTARTSYS;
+ }
+
+ spin_lock_irq(&ev_file->lock);
+ if (ev_file->is_destroyed) {
+ spin_unlock_irq(&ev_file->lock);
+ return -EIO;
+ }
+ }
+
+ if (omit_data) {
+ event_sub = list_first_entry(&ev_file->event_list,
+ struct devx_event_subscription,
+ event_list);
+ eventsz = sizeof(event_sub->cookie);
+ event_data = &event_sub->cookie;
+ } else {
+ event = list_first_entry(&ev_file->event_list,
+ struct devx_async_event_data, list);
+ eventsz = sizeof(struct mlx5_eqe) +
+ sizeof(struct mlx5_ib_uapi_devx_async_event_hdr);
+ event_data = &event->hdr;
+ }
+
+ if (eventsz > count) {
+ spin_unlock_irq(&ev_file->lock);
+ return -EINVAL;
+ }
+
+ if (omit_data)
+ list_del_init(&event_sub->event_list);
+ else
+ list_del(&event->list);
+
+ spin_unlock_irq(&ev_file->lock);
+
+ if (copy_to_user(buf, event_data, eventsz))
+ /* This points to an application issue, not a kernel concern */
+ ret = -EFAULT;
+ else
+ ret = eventsz;
+
+ if (!omit_data)
+ kfree(event);
+ return ret;
}
static __poll_t devx_async_event_poll(struct file *filp,
struct poll_table_struct *wait)
{
- return 0;
+ struct devx_async_event_file *ev_file = filp->private_data;
+ __poll_t pollflags = 0;
+
+ poll_wait(filp, &ev_file->poll_wait, wait);
+
+ spin_lock_irq(&ev_file->lock);
+ if (ev_file->is_destroyed)
+ pollflags = EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
+ else if (!list_empty(&ev_file->event_list))
+ pollflags = EPOLLIN | EPOLLRDNORM;
+ spin_unlock_irq(&ev_file->lock);
+
+ return pollflags;
}
static int devx_async_event_close(struct inode *inode, struct file *filp)
{
struct devx_async_event_file *ev_file = filp->private_data;
struct devx_event_subscription *event_sub, *event_sub_tmp;
+ struct devx_async_event_data *entry, *tmp;
mutex_lock(&ev_file->dev->devx_event_table.event_xa_lock);
/* delete the subscriptions which are related to this FD */
@@ -2338,6 +2617,15 @@ static int devx_async_event_close(struct inode *inode, struct file *filp)
mutex_unlock(&ev_file->dev->devx_event_table.event_xa_lock);
+ /* free the pending events allocation */
+ if (!ev_file->omit_data) {
+ spin_lock_irq(&ev_file->lock);
+ list_for_each_entry_safe(entry, tmp,
+ &ev_file->event_list, list)
+ kfree(entry); /* read can't come any more */
+ spin_unlock_irq(&ev_file->lock);
+ }
+
uverbs_close_fd(filp);
put_device(&ev_file->dev->ib_dev.dev);
return 0;
@@ -2373,6 +2661,15 @@ static int devx_hot_unplug_async_cmd_event_file(struct ib_uobject *uobj,
static int devx_hot_unplug_async_event_file(struct ib_uobject *uobj,
enum rdma_remove_reason why)
{
+ struct devx_async_event_file *ev_file =
+ container_of(uobj, struct devx_async_event_file,
+ uobj);
+
+ spin_lock_irq(&ev_file->lock);
+ ev_file->is_destroyed = 1;
+ spin_unlock_irq(&ev_file->lock);
+
+ wake_up_interruptible(&ev_file->poll_wait);
return 0;
};
diff --git a/include/uapi/rdma/mlx5_user_ioctl_verbs.h b/include/uapi/rdma/mlx5_user_ioctl_verbs.h
index b44691315d39..7e9900b0e746 100644
--- a/include/uapi/rdma/mlx5_user_ioctl_verbs.h
+++ b/include/uapi/rdma/mlx5_user_ioctl_verbs.h
@@ -67,5 +67,10 @@ enum mlx5_ib_uapi_devx_create_event_channel_flags {
MLX5_IB_UAPI_DEVX_CR_EV_CH_FLAGS_OMIT_DATA = 1 << 0,
};
+struct mlx5_ib_uapi_devx_async_event_hdr {
+ __aligned_u64 cookie;
+ __u8 out_data[];
+};
+
#endif
--
2.20.1
^ permalink raw reply related
* [PATCH rdma-next v2 12/13] IB/mlx5: Add DEVX support for CQ events
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Add DEVX support for CQ events by creating and destroying the CQ via
mlx5_core and set an handler to manage its completions.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/devx.c | 39 +++++++++++++++++++++++++++++++
1 file changed, 39 insertions(+)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index de068fe94363..38064f4bfdf6 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -19,9 +19,12 @@
#define UVERBS_MODULE_NAME mlx5_ib
#include <rdma/uverbs_named_ioctl.h>
+static void dispatch_event_fd(struct list_head *fd_list, const void *data);
+
enum devx_obj_flags {
DEVX_OBJ_FLAGS_INDIRECT_MKEY = 1 << 0,
DEVX_OBJ_FLAGS_DCT = 1 << 1,
+ DEVX_OBJ_FLAGS_CQ = 1 << 2,
};
struct devx_async_data {
@@ -89,6 +92,7 @@ struct devx_async_event_file {
#define MLX5_MAX_DESTROY_INBOX_SIZE_DW MLX5_ST_SZ_DW(delete_fte_in)
struct devx_obj {
struct mlx5_core_dev *mdev;
+ struct mlx5_ib_dev *ib_dev;
u64 obj_id;
u32 dinlen; /* destroy inbox length */
u32 dinbox[MLX5_MAX_DESTROY_INBOX_SIZE_DW];
@@ -96,6 +100,7 @@ struct devx_obj {
union {
struct mlx5_ib_devx_mr devx_mr;
struct mlx5_core_dct core_dct;
+ struct mlx5_core_cq core_cq;
};
struct list_head event_sub; /* holds devx_event_subscription entries */
};
@@ -1346,6 +1351,8 @@ static int devx_obj_cleanup(struct ib_uobject *uobject,
if (obj->flags & DEVX_OBJ_FLAGS_DCT)
ret = mlx5_core_destroy_dct(obj->mdev, &obj->core_dct);
+ else if (obj->flags & DEVX_OBJ_FLAGS_CQ)
+ ret = mlx5_core_destroy_cq(obj->mdev, &obj->core_cq);
else
ret = mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out,
sizeof(out));
@@ -1369,6 +1376,29 @@ static int devx_obj_cleanup(struct ib_uobject *uobject,
return ret;
}
+static void devx_cq_comp(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe)
+{
+ struct devx_obj *obj = container_of(mcq, struct devx_obj, core_cq);
+ struct mlx5_devx_event_table *table;
+ struct devx_event *event;
+ struct devx_obj_event *obj_event;
+ u32 obj_id = mcq->cqn;
+
+ table = &obj->ib_dev->devx_event_table;
+ rcu_read_lock();
+ event = xa_load(&table->event_xa, MLX5_EVENT_TYPE_COMP);
+ if (!event)
+ goto out;
+
+ obj_event = xa_load(&event->object_ids, obj_id);
+ if (!obj_event)
+ goto out;
+
+ dispatch_event_fd(&obj_event->obj_sub_list, eqe);
+out:
+ rcu_read_unlock();
+}
+
static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
struct uverbs_attr_bundle *attrs)
{
@@ -1420,6 +1450,12 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
err = mlx5_core_create_dct(dev->mdev, &obj->core_dct,
cmd_in, cmd_in_len,
cmd_out, cmd_out_len);
+ } else if (opcode == MLX5_CMD_OP_CREATE_CQ) {
+ obj->flags |= DEVX_OBJ_FLAGS_CQ;
+ obj->core_cq.comp = devx_cq_comp;
+ err = mlx5_core_create_cq(dev->mdev, &obj->core_cq,
+ cmd_in, cmd_in_len, cmd_out,
+ cmd_out_len);
} else {
err = mlx5_cmd_exec(dev->mdev, cmd_in,
cmd_in_len,
@@ -1432,6 +1468,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
uobj->object = obj;
obj->mdev = dev->mdev;
INIT_LIST_HEAD(&obj->event_sub);
+ obj->ib_dev = dev;
devx_obj_build_destroy_cmd(cmd_in, cmd_out, obj->dinbox, &obj->dinlen,
&obj_id);
WARN_ON(obj->dinlen > MLX5_MAX_DESTROY_INBOX_SIZE_DW * sizeof(u32));
@@ -1459,6 +1496,8 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
obj_destroy:
if (obj->flags & DEVX_OBJ_FLAGS_DCT)
mlx5_core_destroy_dct(obj->mdev, &obj->core_dct);
+ else if (obj->flags & DEVX_OBJ_FLAGS_CQ)
+ mlx5_core_destroy_cq(obj->mdev, &obj->core_cq);
else
mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out,
sizeof(out));
--
2.20.1
^ permalink raw reply related
* [PATCH rdma-next v2 13/13] IB/mlx5: DEVX cleanup mdev
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
No need any more to hold mlx5_core_dev on the devx_object, it can be
accessed from ib_dev.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/devx.c | 18 ++++++++----------
1 file changed, 8 insertions(+), 10 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index 38064f4bfdf6..cb4f0fc79176 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -91,7 +91,6 @@ struct devx_async_event_file {
#define MLX5_MAX_DESTROY_INBOX_SIZE_DW MLX5_ST_SZ_DW(delete_fte_in)
struct devx_obj {
- struct mlx5_core_dev *mdev;
struct mlx5_ib_dev *ib_dev;
u64 obj_id;
u32 dinlen; /* destroy inbox length */
@@ -1297,7 +1296,7 @@ static void devx_free_indirect_mkey(struct rcu_head *rcu)
*/
static void devx_cleanup_mkey(struct devx_obj *obj)
{
- struct mlx5_mkey_table *table = &obj->mdev->priv.mkey_table;
+ struct mlx5_mkey_table *table = &obj->ib_dev->mdev->priv.mkey_table;
unsigned long flags;
write_lock_irqsave(&table->lock, flags);
@@ -1350,12 +1349,12 @@ static int devx_obj_cleanup(struct ib_uobject *uobject,
devx_cleanup_mkey(obj);
if (obj->flags & DEVX_OBJ_FLAGS_DCT)
- ret = mlx5_core_destroy_dct(obj->mdev, &obj->core_dct);
+ ret = mlx5_core_destroy_dct(obj->ib_dev->mdev, &obj->core_dct);
else if (obj->flags & DEVX_OBJ_FLAGS_CQ)
- ret = mlx5_core_destroy_cq(obj->mdev, &obj->core_cq);
+ ret = mlx5_core_destroy_cq(obj->ib_dev->mdev, &obj->core_cq);
else
- ret = mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out,
- sizeof(out));
+ ret = mlx5_cmd_exec(obj->ib_dev->mdev, obj->dinbox,
+ obj->dinlen, out, sizeof(out));
if (ib_is_destroy_retryable(ret, why, uobject))
return ret;
@@ -1466,7 +1465,6 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
goto obj_free;
uobj->object = obj;
- obj->mdev = dev->mdev;
INIT_LIST_HEAD(&obj->event_sub);
obj->ib_dev = dev;
devx_obj_build_destroy_cmd(cmd_in, cmd_out, obj->dinbox, &obj->dinlen,
@@ -1495,11 +1493,11 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
devx_cleanup_mkey(obj);
obj_destroy:
if (obj->flags & DEVX_OBJ_FLAGS_DCT)
- mlx5_core_destroy_dct(obj->mdev, &obj->core_dct);
+ mlx5_core_destroy_dct(obj->ib_dev->mdev, &obj->core_dct);
else if (obj->flags & DEVX_OBJ_FLAGS_CQ)
- mlx5_core_destroy_cq(obj->mdev, &obj->core_cq);
+ mlx5_core_destroy_cq(obj->ib_dev->mdev, &obj->core_cq);
else
- mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out,
+ mlx5_cmd_exec(obj->ib_dev->mdev, obj->dinbox, obj->dinlen, out,
sizeof(out));
obj_free:
kfree(obj);
--
2.20.1
^ permalink raw reply related
* [PATCH rdma-next v2 09/13] IB/mlx5: Register DEVX with mlx5_core to get async events
From: Leon Romanovsky @ 2019-06-30 16:23 UTC (permalink / raw)
To: Doug Ledford, Jason Gunthorpe
Cc: Leon Romanovsky, RDMA mailing list, Saeed Mahameed, Yishai Hadas,
linux-netdev
In-Reply-To: <20190630162334.22135-1-leon@kernel.org>
From: Yishai Hadas <yishaih@mellanox.com>
Register DEVX with with mlx5_core to get async events. This will enable
to dispatch the applicable events to its consumers in down stream
patches.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
drivers/infiniband/hw/mlx5/devx.c | 30 ++++++++++++++++++++++++++++
drivers/infiniband/hw/mlx5/main.c | 8 ++++++--
drivers/infiniband/hw/mlx5/mlx5_ib.h | 12 +++++++++++
3 files changed, 48 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index 89947b4f4f6f..b7b17e882d82 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -1673,6 +1673,36 @@ static int devx_umem_cleanup(struct ib_uobject *uobject,
return 0;
}
+static int devx_event_notifier(struct notifier_block *nb,
+ unsigned long event_type, void *data)
+{
+ return NOTIFY_DONE;
+}
+
+void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev)
+{
+ struct mlx5_devx_event_table *table = &dev->devx_event_table;
+
+ xa_init(&table->event_xa);
+ mutex_init(&table->event_xa_lock);
+ MLX5_NB_INIT(&table->devx_nb, devx_event_notifier, NOTIFY_ANY);
+ mlx5_eq_notifier_register(dev->mdev, &table->devx_nb);
+}
+
+void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev)
+{
+ struct mlx5_devx_event_table *table = &dev->devx_event_table;
+ void *entry;
+ unsigned long id;
+
+ mlx5_eq_notifier_unregister(dev->mdev, &table->devx_nb);
+
+ xa_for_each(&table->event_xa, id, entry)
+ kfree(entry);
+
+ xa_destroy(&table->event_xa);
+}
+
static ssize_t devx_async_cmd_event_read(struct file *filp, char __user *buf,
size_t count, loff_t *pos)
{
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 23997f744586..be0e04cfed77 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -6652,15 +6652,19 @@ static int mlx5_ib_stage_devx_init(struct mlx5_ib_dev *dev)
int uid;
uid = mlx5_ib_devx_create(dev, false);
- if (uid > 0)
+ if (uid > 0) {
dev->devx_whitelist_uid = uid;
+ mlx5_ib_devx_init_event_table(dev);
+ }
return 0;
}
static void mlx5_ib_stage_devx_cleanup(struct mlx5_ib_dev *dev)
{
- if (dev->devx_whitelist_uid)
+ if (dev->devx_whitelist_uid) {
+ mlx5_ib_devx_cleanup_event_table(dev);
mlx5_ib_devx_destroy(dev, dev->devx_whitelist_uid);
+ }
}
void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 9cf23ae6324e..556af34b788b 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -944,6 +944,13 @@ struct mlx5_ib_pf_eq {
mempool_t *pool;
};
+struct mlx5_devx_event_table {
+ struct mlx5_nb devx_nb;
+ /* serialize updating the event_xa */
+ struct mutex event_xa_lock;
+ struct xarray event_xa;
+};
+
struct mlx5_ib_dev {
struct ib_device ib_dev;
struct mlx5_core_dev *mdev;
@@ -994,6 +1001,7 @@ struct mlx5_ib_dev {
struct mlx5_srq_table srq_table;
struct mlx5_async_ctx async_ctx;
int free_port;
+ struct mlx5_devx_event_table devx_event_table;
};
static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
@@ -1333,6 +1341,8 @@ void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user);
void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid);
+void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev);
+void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev);
const struct uverbs_object_tree_def *mlx5_ib_get_devx_tree(void);
extern const struct uapi_definition mlx5_ib_devx_defs[];
extern const struct uapi_definition mlx5_ib_flow_defs[];
@@ -1349,6 +1359,8 @@ static inline int
mlx5_ib_devx_create(struct mlx5_ib_dev *dev,
bool is_user) { return -EOPNOTSUPP; }
static inline void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid) {}
+static inline void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev) {}
+static inline void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev) {}
static inline bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id,
int *dest_type)
{
--
2.20.1
^ permalink raw reply related
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ivan Khoronzhuk @ 2019-06-30 16:25 UTC (permalink / raw)
To: Ilias Apalodimas
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <1561785805-21647-4-git-send-email-ilias.apalodimas@linaro.org>
On Sat, Jun 29, 2019 at 08:23:25AM +0300, Ilias Apalodimas wrote:
>The interface only supports 1 Tx queue so locking is introduced on
>the Tx queue if XDP is enabled to make sure .ndo_start_xmit and
>.ndo_xdp_xmit won't corrupt Tx ring
>
>- Performance (SMMU off)
>
>Benchmark XDP_SKB XDP_DRV
>xdp1 291kpps 344kpps
>rxdrop 282kpps 342kpps
>
>- Performance (SMMU on)
>Benchmark XDP_SKB XDP_DRV
>xdp1 167kpps 324kpps
>rxdrop 164kpps 323kpps
>
>Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
>---
> drivers/net/ethernet/socionext/netsec.c | 361 ++++++++++++++++++++++--
> 1 file changed, 334 insertions(+), 27 deletions(-)
>
[...]
>+
>+static int netsec_xdp_setup(struct netsec_priv *priv, struct bpf_prog *prog,
>+ struct netlink_ext_ack *extack)
>+{
>+ struct net_device *dev = priv->ndev;
>+ struct bpf_prog *old_prog;
>+
>+ /* For now just support only the usual MTU sized frames */
>+ if (prog && dev->mtu > 1500) {
>+ NL_SET_ERR_MSG_MOD(extack, "Jumbo frames not supported on XDP");
>+ return -EOPNOTSUPP;
>+ }
>+
>+ if (netif_running(dev))
>+ netsec_netdev_stop(dev);
And why to stop the interface. XDP allows to update prog in runtime.
>+
>+ /* Detach old prog, if any */
>+ old_prog = xchg(&priv->xdp_prog, prog);
>+ if (old_prog)
>+ bpf_prog_put(old_prog);
>+
>+ if (netif_running(dev))
>+ netsec_netdev_open(dev);
>+
>+ return 0;
>+}
>+
>+static int netsec_xdp(struct net_device *ndev, struct netdev_bpf *xdp)
>+{
>+ struct netsec_priv *priv = netdev_priv(ndev);
>+
>+ switch (xdp->command) {
>+ case XDP_SETUP_PROG:
>+ return netsec_xdp_setup(priv, xdp->prog, xdp->extack);
>+ case XDP_QUERY_PROG:
>+ xdp->prog_id = priv->xdp_prog ? priv->xdp_prog->aux->id : 0;
>+ return 0;
>+ default:
>+ return -EINVAL;
>+ }
>+}
>+
> static const struct net_device_ops netsec_netdev_ops = {
> .ndo_init = netsec_netdev_init,
> .ndo_uninit = netsec_netdev_uninit,
>@@ -1537,6 +1842,8 @@ static const struct net_device_ops netsec_netdev_ops = {
> .ndo_set_mac_address = eth_mac_addr,
> .ndo_validate_addr = eth_validate_addr,
> .ndo_do_ioctl = netsec_netdev_ioctl,
>+ .ndo_xdp_xmit = netsec_xdp_xmit,
>+ .ndo_bpf = netsec_xdp,
> };
>
> static int netsec_of_probe(struct platform_device *pdev,
>--
>2.20.1
>
--
Regards,
Ivan Khoronzhuk
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ilias Apalodimas @ 2019-06-30 16:32 UTC (permalink / raw)
To: Ivan Khoronzhuk
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630162552.GB12704@khorivan>
On Sun, Jun 30, 2019 at 07:25:53PM +0300, Ivan Khoronzhuk wrote:
> On Sat, Jun 29, 2019 at 08:23:25AM +0300, Ilias Apalodimas wrote:
> >The interface only supports 1 Tx queue so locking is introduced on
> >the Tx queue if XDP is enabled to make sure .ndo_start_xmit and
> >.ndo_xdp_xmit won't corrupt Tx ring
> >
> >- Performance (SMMU off)
> >
> >Benchmark XDP_SKB XDP_DRV
> >xdp1 291kpps 344kpps
> >rxdrop 282kpps 342kpps
> >
> >- Performance (SMMU on)
> >Benchmark XDP_SKB XDP_DRV
> >xdp1 167kpps 324kpps
> >rxdrop 164kpps 323kpps
> >
> >Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
> >---
> >drivers/net/ethernet/socionext/netsec.c | 361 ++++++++++++++++++++++--
> >1 file changed, 334 insertions(+), 27 deletions(-)
> >
>
> [...]
>
> >+
> >+static int netsec_xdp_setup(struct netsec_priv *priv, struct bpf_prog *prog,
> >+ struct netlink_ext_ack *extack)
> >+{
> >+ struct net_device *dev = priv->ndev;
> >+ struct bpf_prog *old_prog;
> >+
> >+ /* For now just support only the usual MTU sized frames */
> >+ if (prog && dev->mtu > 1500) {
> >+ NL_SET_ERR_MSG_MOD(extack, "Jumbo frames not supported on XDP");
> >+ return -EOPNOTSUPP;
> >+ }
> >+
> >+ if (netif_running(dev))
> >+ netsec_netdev_stop(dev);
> And why to stop the interface. XDP allows to update prog in runtime.
>
Adding the support is not limited to adding a prog only in this driver.
It also rebuilts the queues which changes the dma mapping of buffers.
Since i don't want to map BIDIRECTIONAL buffers if XDP is not in place,
i am resetting the device and forcing the buffer re-allocation
Thanks
/Ilias
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ilias Apalodimas @ 2019-06-30 16:34 UTC (permalink / raw)
To: Ivan Khoronzhuk
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630162042.GA12704@khorivan>
Hi Ivan,
>
> [...]
>
> >+
> >+static int netsec_xdp(struct net_device *ndev, struct netdev_bpf *xdp)
> >+{
> >+ struct netsec_priv *priv = netdev_priv(ndev);
> >+
> >+ switch (xdp->command) {
> >+ case XDP_SETUP_PROG:
> >+ return netsec_xdp_setup(priv, xdp->prog, xdp->extack);
> >+ case XDP_QUERY_PROG:
> >+ xdp->prog_id = priv->xdp_prog ? priv->xdp_prog->aux->id : 0;
> xdp_attachment family to save bpf flags?
Sure why not. This can always be added later though since many drivers are
already doing it similarly no?
>
> >+ return 0;
> >+ default:
> >+ return -EINVAL;
> >+ }
> >+}
> >+
> >static const struct net_device_ops netsec_netdev_ops = {
> > .ndo_init = netsec_netdev_init,
> > .ndo_uninit = netsec_netdev_uninit,
> >@@ -1537,6 +1842,8 @@ static const struct net_device_ops netsec_netdev_ops = {
> > .ndo_set_mac_address = eth_mac_addr,
> > .ndo_validate_addr = eth_validate_addr,
> > .ndo_do_ioctl = netsec_netdev_ioctl,
> >+ .ndo_xdp_xmit = netsec_xdp_xmit,
> >+ .ndo_bpf = netsec_xdp,
> >};
> >
> >static int netsec_of_probe(struct platform_device *pdev,
> >--
> >2.20.1
> >
>
> --
> Regards,
> Ivan Khoronzhuk
Thanks
/Ilias
^ permalink raw reply
* Re: [PATCH net v3] vxlan: do not destroy fdb if register_netdevice() is failed
From: Roopa Prabhu @ 2019-06-30 16:40 UTC (permalink / raw)
To: Taehee Yoo; +Cc: David Miller, Petr Machata, netdev
In-Reply-To: <20190628050725.9445-1-ap420073@gmail.com>
On Thu, Jun 27, 2019 at 10:07 PM Taehee Yoo <ap420073@gmail.com> wrote:
>
> __vxlan_dev_create() destroys FDB using specific pointer which indicates
> a fdb when error occurs.
> But that pointer should not be used when register_netdevice() fails because
> register_netdevice() internally destroys fdb when error occurs.
>
> This patch makes vxlan_fdb_create() to do not link fdb entry to vxlan dev
> internally.
> Instead, a new function vxlan_fdb_insert() is added to link fdb to vxlan
> dev.
>
> vxlan_fdb_insert() is called after calling register_netdevice().
> This routine can avoid situation that ->ndo_uninit() destroys fdb entry
> in error path of register_netdevice().
> Hence, error path of __vxlan_dev_create() routine can have an opportunity
> to destroy default fdb entry by hand.
>
> Test command
> ip link add bonding_masters type vxlan id 0 group 239.1.1.1 \
> dev enp0s9 dstport 4789
>
> Splat looks like:
> [ 213.392816] kasan: GPF could be caused by NULL-ptr deref or user memory access
> [ 213.401257] general protection fault: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI
> [ 213.402178] CPU: 0 PID: 1414 Comm: ip Not tainted 5.2.0-rc5+ #256
> [ 213.402178] RIP: 0010:vxlan_fdb_destroy+0x120/0x220 [vxlan]
> [ 213.402178] Code: df 48 8b 2b 48 89 fa 48 c1 ea 03 80 3c 02 00 0f 85 06 01 00 00 4c 8b 63 08 48 b8 00 00 00 00 00 fc d
> [ 213.402178] RSP: 0018:ffff88810cb9f0a0 EFLAGS: 00010202
> [ 213.402178] RAX: dffffc0000000000 RBX: ffff888101d4a8c8 RCX: 0000000000000000
> [ 213.402178] RDX: 1bd5a00000000040 RSI: ffff888101d4a8c8 RDI: ffff888101d4a8d0
> [ 213.402178] RBP: 0000000000000000 R08: fffffbfff22b72d9 R09: 0000000000000000
> [ 213.402178] R10: 00000000ffffffef R11: 0000000000000000 R12: dead000000000200
> [ 213.402178] R13: ffff88810cb9f1f8 R14: ffff88810efccda0 R15: ffff88810efccda0
> [ 213.402178] FS: 00007f7f6621a0c0(0000) GS:ffff88811b000000(0000) knlGS:0000000000000000
> [ 213.402178] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 213.402178] CR2: 000055746f0807d0 CR3: 00000001123e0000 CR4: 00000000001006f0
> [ 213.402178] Call Trace:
> [ 213.402178] __vxlan_dev_create+0x3a9/0x7d0 [vxlan]
> [ 213.402178] ? vxlan_changelink+0x740/0x740 [vxlan]
> [ 213.402178] ? rcu_read_unlock+0x60/0x60 [vxlan]
> [ 213.402178] ? __kasan_kmalloc.constprop.3+0xa0/0xd0
> [ 213.402178] vxlan_newlink+0x8d/0xc0 [vxlan]
> [ 213.402178] ? __vxlan_dev_create+0x7d0/0x7d0 [vxlan]
> [ 213.554119] ? __netlink_ns_capable+0xc3/0xf0
> [ 213.554119] __rtnl_newlink+0xb75/0x1180
> [ 213.554119] ? rtnl_link_unregister+0x230/0x230
> [ ... ]
>
> Fixes: 0241b836732f ("vxlan: fix default fdb entry netlink notify ordering during netdev create")
> Suggested-by: Roopa Prabhu <roopa@cumulusnetworks.com>
> Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
LGTM, thanks!.
> ---
>
> v2 -> v3 :
> - Use function name vxlan_fdb_insert() instead of vxlan_fdb_link()
> - Add and use __vxlan_fdb_free() instead of call_rcu()
>
> v1 -> v2 :
> - Add a new function vxlan_fdb_link().
> - Fix fdb entry leak.
> - Update description.
>
> drivers/net/vxlan.c | 37 +++++++++++++++++++++++++++----------
> 1 file changed, 27 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
> index 083f3f0bf37f..b4283f52a09d 100644
> --- a/drivers/net/vxlan.c
> +++ b/drivers/net/vxlan.c
> @@ -804,6 +804,14 @@ static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
> return f;
> }
>
> +static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
> + __be32 src_vni, struct vxlan_fdb *f)
> +{
> + ++vxlan->addrcnt;
> + hlist_add_head_rcu(&f->hlist,
> + vxlan_fdb_head(vxlan, mac, src_vni));
> +}
> +
> static int vxlan_fdb_create(struct vxlan_dev *vxlan,
> const u8 *mac, union vxlan_addr *ip,
> __u16 state, __be16 port, __be32 src_vni,
> @@ -829,18 +837,13 @@ static int vxlan_fdb_create(struct vxlan_dev *vxlan,
> return rc;
> }
>
> - ++vxlan->addrcnt;
> - hlist_add_head_rcu(&f->hlist,
> - vxlan_fdb_head(vxlan, mac, src_vni));
> -
> *fdb = f;
>
> return 0;
> }
>
> -static void vxlan_fdb_free(struct rcu_head *head)
> +static void __vxlan_fdb_free(struct vxlan_fdb *f)
> {
> - struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
> struct vxlan_rdst *rd, *nd;
>
> list_for_each_entry_safe(rd, nd, &f->remotes, list) {
> @@ -850,6 +853,13 @@ static void vxlan_fdb_free(struct rcu_head *head)
> kfree(f);
> }
>
> +static void vxlan_fdb_free(struct rcu_head *head)
> +{
> + struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
> +
> + __vxlan_fdb_free(f);
> +}
> +
> static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
> bool do_notify, bool swdev_notify)
> {
> @@ -977,6 +987,7 @@ static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
> if (rc < 0)
> return rc;
>
> + vxlan_fdb_insert(vxlan, mac, src_vni, f);
> rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
> swdev_notify, extack);
> if (rc)
> @@ -3571,12 +3582,17 @@ static int __vxlan_dev_create(struct net *net, struct net_device *dev,
> if (err)
> goto errout;
>
> - /* notify default fdb entry */
> if (f) {
> + vxlan_fdb_insert(vxlan, all_zeros_mac,
> + vxlan->default_dst.remote_vni, f);
> +
> + /* notify default fdb entry */
> err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
> RTM_NEWNEIGH, true, extack);
> - if (err)
> - goto errout;
> + if (err) {
> + vxlan_fdb_destroy(vxlan, f, false, false);
> + goto unregister;
> + }
> }
>
> list_add(&vxlan->next, &vn->vxlan_list);
> @@ -3588,7 +3604,8 @@ static int __vxlan_dev_create(struct net *net, struct net_device *dev,
> * destroy the entry by hand here.
> */
> if (f)
> - vxlan_fdb_destroy(vxlan, f, false, false);
> + __vxlan_fdb_free(f);
> +unregister:
> if (unregister)
> unregister_netdevice(dev);
> return err;
> --
> 2.17.1
>
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ivan Khoronzhuk @ 2019-06-30 16:41 UTC (permalink / raw)
To: Ilias Apalodimas
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630163214.GA10484@apalos>
On Sun, Jun 30, 2019 at 07:32:14PM +0300, Ilias Apalodimas wrote:
>On Sun, Jun 30, 2019 at 07:25:53PM +0300, Ivan Khoronzhuk wrote:
>> On Sat, Jun 29, 2019 at 08:23:25AM +0300, Ilias Apalodimas wrote:
>> >The interface only supports 1 Tx queue so locking is introduced on
>> >the Tx queue if XDP is enabled to make sure .ndo_start_xmit and
>> >.ndo_xdp_xmit won't corrupt Tx ring
>> >
>> >- Performance (SMMU off)
>> >
>> >Benchmark XDP_SKB XDP_DRV
>> >xdp1 291kpps 344kpps
>> >rxdrop 282kpps 342kpps
>> >
>> >- Performance (SMMU on)
>> >Benchmark XDP_SKB XDP_DRV
>> >xdp1 167kpps 324kpps
>> >rxdrop 164kpps 323kpps
>> >
>> >Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
>> >---
>> >drivers/net/ethernet/socionext/netsec.c | 361 ++++++++++++++++++++++--
>> >1 file changed, 334 insertions(+), 27 deletions(-)
>> >
>>
>> [...]
>>
>> >+
>> >+static int netsec_xdp_setup(struct netsec_priv *priv, struct bpf_prog *prog,
>> >+ struct netlink_ext_ack *extack)
>> >+{
>> >+ struct net_device *dev = priv->ndev;
>> >+ struct bpf_prog *old_prog;
>> >+
>> >+ /* For now just support only the usual MTU sized frames */
>> >+ if (prog && dev->mtu > 1500) {
>> >+ NL_SET_ERR_MSG_MOD(extack, "Jumbo frames not supported on XDP");
>> >+ return -EOPNOTSUPP;
>> >+ }
>> >+
>> >+ if (netif_running(dev))
>> >+ netsec_netdev_stop(dev);
>> And why to stop the interface. XDP allows to update prog in runtime.
>>
>Adding the support is not limited to adding a prog only in this driver.
>It also rebuilts the queues which changes the dma mapping of buffers.
>Since i don't want to map BIDIRECTIONAL buffers if XDP is not in place,
>i am resetting the device and forcing the buffer re-allocation
>
>Thanks
>/Ilias
I don't know the internals, probably it has some dependencies, but here you
just update the prog and can at least do it when exchange is happening.
I mean not in case of prog is attached/removed first time.
In case of prog -> prog it seems doable...
It ups to you ofc, but I can run smth like:
ip -force link set dev eth0 xdp obj xdp-example-pass.o sec .text
and expect it's updated w/o interface reset I mean on new prog.
I'm not sure, but maintainers can help, conceptually it's supposed to be in
runtime the prog be update uder rcu as a part of API usage...
--
Regards,
Ivan Khoronzhuk
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ivan Khoronzhuk @ 2019-06-30 16:45 UTC (permalink / raw)
To: Ilias Apalodimas
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630163417.GB10484@apalos>
On Sun, Jun 30, 2019 at 07:34:17PM +0300, Ilias Apalodimas wrote:
>Hi Ivan,
>>
>> [...]
>>
>> >+
>> >+static int netsec_xdp(struct net_device *ndev, struct netdev_bpf *xdp)
>> >+{
>> >+ struct netsec_priv *priv = netdev_priv(ndev);
>> >+
>> >+ switch (xdp->command) {
>> >+ case XDP_SETUP_PROG:
>> >+ return netsec_xdp_setup(priv, xdp->prog, xdp->extack);
>> >+ case XDP_QUERY_PROG:
>> >+ xdp->prog_id = priv->xdp_prog ? priv->xdp_prog->aux->id : 0;
>> xdp_attachment family to save bpf flags?
>Sure why not. This can always be added later though since many drivers are
>already doing it similarly no?
yes.
I can work w/o this ofc.
But netronome and cpsw (me) added this.
What I've seen it allows to prevent prog update if flag doesn't allow it.
Usually it doesn't allow, but can be forced with flag. In another case it can
be updated any time w/o reason...and seems like in your case it's sensitive.
>
>>
>> >+ return 0;
>> >+ default:
>> >+ return -EINVAL;
>> >+ }
>> >+}
>> >+
>> >static const struct net_device_ops netsec_netdev_ops = {
>> > .ndo_init = netsec_netdev_init,
>> > .ndo_uninit = netsec_netdev_uninit,
>> >@@ -1537,6 +1842,8 @@ static const struct net_device_ops netsec_netdev_ops = {
>> > .ndo_set_mac_address = eth_mac_addr,
>> > .ndo_validate_addr = eth_validate_addr,
>> > .ndo_do_ioctl = netsec_netdev_ioctl,
>> >+ .ndo_xdp_xmit = netsec_xdp_xmit,
>> >+ .ndo_bpf = netsec_xdp,
>> >};
>> >
>> >static int netsec_of_probe(struct platform_device *pdev,
>> >--
>> >2.20.1
>> >
>>
>> --
>> Regards,
>> Ivan Khoronzhuk
>
>Thanks
>/Ilias
--
Regards,
Ivan Khoronzhuk
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ilias Apalodimas @ 2019-06-30 16:47 UTC (permalink / raw)
To: Ivan Khoronzhuk
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630164127.GC12704@khorivan>
On Sun, Jun 30, 2019 at 07:41:28PM +0300, Ivan Khoronzhuk wrote:
> On Sun, Jun 30, 2019 at 07:32:14PM +0300, Ilias Apalodimas wrote:
> >On Sun, Jun 30, 2019 at 07:25:53PM +0300, Ivan Khoronzhuk wrote:
> >>On Sat, Jun 29, 2019 at 08:23:25AM +0300, Ilias Apalodimas wrote:
> >>>The interface only supports 1 Tx queue so locking is introduced on
> >>>the Tx queue if XDP is enabled to make sure .ndo_start_xmit and
> >>>.ndo_xdp_xmit won't corrupt Tx ring
> >>>
> >>>- Performance (SMMU off)
> >>>
> >>>Benchmark XDP_SKB XDP_DRV
> >>>xdp1 291kpps 344kpps
> >>>rxdrop 282kpps 342kpps
> >>>
> >>>- Performance (SMMU on)
> >>>Benchmark XDP_SKB XDP_DRV
> >>>xdp1 167kpps 324kpps
> >>>rxdrop 164kpps 323kpps
> >>>
> >>>Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
> >>>---
> >>>drivers/net/ethernet/socionext/netsec.c | 361 ++++++++++++++++++++++--
> >>>1 file changed, 334 insertions(+), 27 deletions(-)
> >>>
> >>
> >>[...]
> >>
> >>>+
> >>>+static int netsec_xdp_setup(struct netsec_priv *priv, struct bpf_prog *prog,
> >>>+ struct netlink_ext_ack *extack)
> >>>+{
> >>>+ struct net_device *dev = priv->ndev;
> >>>+ struct bpf_prog *old_prog;
> >>>+
> >>>+ /* For now just support only the usual MTU sized frames */
> >>>+ if (prog && dev->mtu > 1500) {
> >>>+ NL_SET_ERR_MSG_MOD(extack, "Jumbo frames not supported on XDP");
> >>>+ return -EOPNOTSUPP;
> >>>+ }
> >>>+
> >>>+ if (netif_running(dev))
> >>>+ netsec_netdev_stop(dev);
> >>And why to stop the interface. XDP allows to update prog in runtime.
> >>
> >Adding the support is not limited to adding a prog only in this driver.
> >It also rebuilts the queues which changes the dma mapping of buffers.
> >Since i don't want to map BIDIRECTIONAL buffers if XDP is not in place,
> >i am resetting the device and forcing the buffer re-allocation
> >
> >Thanks
> >/Ilias
> I don't know the internals, probably it has some dependencies, but here you
> just update the prog and can at least do it when exchange is happening.
> I mean not in case of prog is attached/removed first time.
> In case of prog -> prog it seems doable...
>
> It ups to you ofc, but I can run smth like:
> ip -force link set dev eth0 xdp obj xdp-example-pass.o sec .text
> and expect it's updated w/o interface reset I mean on new prog.
>
> I'm not sure, but maintainers can help, conceptually it's supposed to be in
> runtime the prog be update uder rcu as a part of API usage...
It's doable but it means i'd have to change the buffer allocation again. I'd
also prefer mapping FOR_DEVICE only if XDP is not loaded. Most of the drivers do
restart so i'll stick with this for the current version.
Most of the drivers do restart now so i'll stick to that for now.
Thanks
/Ilias
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ilias Apalodimas @ 2019-06-30 16:50 UTC (permalink / raw)
To: Ivan Khoronzhuk
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630164512.GD12704@khorivan>
On Sun, Jun 30, 2019 at 07:45:13PM +0300, Ivan Khoronzhuk wrote:
> On Sun, Jun 30, 2019 at 07:34:17PM +0300, Ilias Apalodimas wrote:
> >Hi Ivan,
> >>
> >>[...]
> >>
> >>>+
> >>>+static int netsec_xdp(struct net_device *ndev, struct netdev_bpf *xdp)
> >>>+{
> >>>+ struct netsec_priv *priv = netdev_priv(ndev);
> >>>+
> >>>+ switch (xdp->command) {
> >>>+ case XDP_SETUP_PROG:
> >>>+ return netsec_xdp_setup(priv, xdp->prog, xdp->extack);
> >>>+ case XDP_QUERY_PROG:
> >>>+ xdp->prog_id = priv->xdp_prog ? priv->xdp_prog->aux->id : 0;
> >>xdp_attachment family to save bpf flags?
> >Sure why not. This can always be added later though since many drivers are
> >already doing it similarly no?
> yes.
> I can work w/o this ofc.
> But netronome and cpsw (me) added this.
Ok let's start using that
> What I've seen it allows to prevent prog update if flag doesn't allow it.
> Usually it doesn't allow, but can be forced with flag. In another case it can
> be updated any time w/o reason...and seems like in your case it's sensitive.
I intend to send a follow up patch anyway to remove the declaration on the top
of the file of netsec_set_tx_de(). I intentionally choose to add that to make
the review easier (since re-arranging would mess that up).
I'll just this optimization as well on the follow up patch since it doesn't
break anything
Thanks
/Ilias
^ permalink raw reply
* Re: [net-next, PATCH 3/3, v2] net: netsec: add XDP support
From: Ivan Khoronzhuk @ 2019-06-30 16:51 UTC (permalink / raw)
To: Ilias Apalodimas
Cc: netdev, jaswinder.singh, ard.biesheuvel, bjorn.topel,
magnus.karlsson, brouer, daniel, ast, makita.toshiaki,
jakub.kicinski, john.fastabend, davem, maciejromanfijalkowski
In-Reply-To: <20190630164745.GA11278@apalos>
On Sun, Jun 30, 2019 at 07:47:45PM +0300, Ilias Apalodimas wrote:
>On Sun, Jun 30, 2019 at 07:41:28PM +0300, Ivan Khoronzhuk wrote:
>> On Sun, Jun 30, 2019 at 07:32:14PM +0300, Ilias Apalodimas wrote:
>> >On Sun, Jun 30, 2019 at 07:25:53PM +0300, Ivan Khoronzhuk wrote:
>> >>On Sat, Jun 29, 2019 at 08:23:25AM +0300, Ilias Apalodimas wrote:
>> >>>The interface only supports 1 Tx queue so locking is introduced on
>> >>>the Tx queue if XDP is enabled to make sure .ndo_start_xmit and
>> >>>.ndo_xdp_xmit won't corrupt Tx ring
>> >>>
>> >>>- Performance (SMMU off)
>> >>>
>> >>>Benchmark XDP_SKB XDP_DRV
>> >>>xdp1 291kpps 344kpps
>> >>>rxdrop 282kpps 342kpps
>> >>>
>> >>>- Performance (SMMU on)
>> >>>Benchmark XDP_SKB XDP_DRV
>> >>>xdp1 167kpps 324kpps
>> >>>rxdrop 164kpps 323kpps
>> >>>
>> >>>Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
>> >>>---
>> >>>drivers/net/ethernet/socionext/netsec.c | 361 ++++++++++++++++++++++--
>> >>>1 file changed, 334 insertions(+), 27 deletions(-)
>> >>>
>> >>
>> >>[...]
>> >>
>> >>>+
>> >>>+static int netsec_xdp_setup(struct netsec_priv *priv, struct bpf_prog *prog,
>> >>>+ struct netlink_ext_ack *extack)
>> >>>+{
>> >>>+ struct net_device *dev = priv->ndev;
>> >>>+ struct bpf_prog *old_prog;
>> >>>+
>> >>>+ /* For now just support only the usual MTU sized frames */
>> >>>+ if (prog && dev->mtu > 1500) {
>> >>>+ NL_SET_ERR_MSG_MOD(extack, "Jumbo frames not supported on XDP");
>> >>>+ return -EOPNOTSUPP;
>> >>>+ }
>> >>>+
>> >>>+ if (netif_running(dev))
>> >>>+ netsec_netdev_stop(dev);
>> >>And why to stop the interface. XDP allows to update prog in runtime.
>> >>
>> >Adding the support is not limited to adding a prog only in this driver.
>> >It also rebuilts the queues which changes the dma mapping of buffers.
>> >Since i don't want to map BIDIRECTIONAL buffers if XDP is not in place,
>> >i am resetting the device and forcing the buffer re-allocation
>> >
>> >Thanks
>> >/Ilias
>> I don't know the internals, probably it has some dependencies, but here you
>> just update the prog and can at least do it when exchange is happening.
>> I mean not in case of prog is attached/removed first time.
>> In case of prog -> prog it seems doable...
>>
>> It ups to you ofc, but I can run smth like:
>> ip -force link set dev eth0 xdp obj xdp-example-pass.o sec .text
>> and expect it's updated w/o interface reset I mean on new prog.
>>
>> I'm not sure, but maintainers can help, conceptually it's supposed to be in
>> runtime the prog be update uder rcu as a part of API usage...
>It's doable but it means i'd have to change the buffer allocation again. I'd
>also prefer mapping FOR_DEVICE only if XDP is not loaded. Most of the drivers do
>restart so i'll stick with this for the current version.
>Most of the drivers do restart now so i'll stick to that for now.
I have nothing against it to be merged. Just pay attention on
XDP_FLAGS_UPDATE_IF_NOEXIST or update it layter or now...ups to your.
>
>Thanks
>/Ilias
--
Regards,
Ivan Khoronzhuk
^ permalink raw reply
* Re: [RFC net-next] net: dsa: add support for MC_DISABLED attribute
From: Linus Lüssing @ 2019-06-30 16:56 UTC (permalink / raw)
To: Ido Schimmel
Cc: Russell King - ARM Linux admin, nikolay, Ido Schimmel,
Vivien Didelot, Florian Fainelli, netdev@vger.kernel.org,
Jiri Pirko, andrew@lunn.ch, davem@davemloft.net, bridge,
b.a.t.m.a.n
In-Reply-To: <20190629162945.GB17143@splinter>
On Sat, Jun 29, 2019 at 07:29:45PM +0300, Ido Schimmel wrote:
> I would like to avoid having drivers take the querier state into account
> as it will only complicate things further.
I absolutely share your pain. Initially in the early prototypes of
multicast awareness in batman-adv we did not consider the querier state.
And doing so later did indeed complicate the code a good bit in batman-adv
(together with the IGMP/MLD suppression issues). I would have loved to
avoid that.
> Is there anything we can do about it? Enable the bridge querier if no
> other querier was detected? Commit c5c23260594c ("bridge: Add
> multicast_querier toggle and disable queries by default") disabled
> queries by default, but I'm only suggesting to turn them on if no other
> querier was detected on the link. Do you think it's still a problem?
As soon as you start becoming the querier, you will not be able to reliably
detect anymore whether you are the only querier candidate.
If any random Linux host using a bridge device were potentially becoming
a querier, that would cause quite some trouble when this host is
behind some bad, bottleneck connection. This host will receive
all multicast traffic, not just IGMP/MLD reports. And with a
congested connection and then unreliable IGMP/MLD, multicast would
become unreliable overall in this domain. So it's important that
your querier is not running in the "dark, remote, dusty closet" of
your network (topologically speaking).
> On Sun, Jun 23, 2019 at 10:44:27AM +0300, Ido Schimmel wrote:
> > See commit b00589af3b04 ("bridge: disable snooping if there is no
> > querier"). I think that's unfortunate behavior that we need because
> > multicast snooping is enabled by default. If it weren't enabled by
> > default, then anyone enabling it would also make sure there's a querier
> > in the network.
I do not quite understand that point. In a way, that's what we
have right now, isn't it? By default it's disabled, because by
default there is no querier on the link. So anyone wanting to use
multicast snooping will need to make sure there's a querier in the
network.
Overall I think the querier (election) mechanism in the standards could
need an update. While the lowest-address first might have
worked well back then, in uniform, fully wired networks where the
position of the querier did not matter, this is not a good
solution anymore in networks involving wireless, dynamic connections.
Especially in wireless mesh networks this is a bit of an issue for
us. Ideally, the querier mechanism were dismissed in favour of simply
unsolicited, periodic IGMP/MLD reports...
But of course, updating IETF standards is no solution for now.
While more complicated, it would not be impossible to consider the
querier state, would it? I mean you probably already need to
consider the case of a user disabling multicast snooping during
runtime, right? So similarly, you could react to appearing or
disappearing queriers?
Cheers, Linus
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox