lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240625120807.1165581-1-amcohen@nvidia.com>
Date: Tue, 25 Jun 2024 15:08:03 +0300
From: Amit Cohen <amcohen@...dia.com>
To: <kuba@...nel.org>
CC: <davem@...emloft.net>, <edumazet@...gle.com>, <pabeni@...hat.com>,
	<hawk@...nel.org>, <idosch@...dia.com>, <petrm@...dia.com>,
	<mlxsw@...dia.com>, <netdev@...r.kernel.org>, Amit Cohen <amcohen@...dia.com>
Subject: [PATCH RFC net-next 0/4] Adjust page pool netlink filling to non common case

Most network drivers has 1:1 mapping between netdevice and event queues,
so then each page pool is used by only one netdevice. This is not the case
in mlxsw driver.

Currently, the netlink message is filled with 'pool->slow.netdev->ifindex',
which should be NULL in case that several netdevices use the same pool.
Adjust page pool netlink filling to use the netdevice which the pool is
stored in its list. See more info in commit messages.

Without this set, mlxsw driver cannot dump all page pools:
$ ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/netdev.yaml \
	--dump page-pool-stats-get --output-json | jq
[]

With this set, "dump" command prints all the page pools for all the
netdevices:
$ ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/netdev.yaml \
	--dump page-pool-get --output-json | \
	jq -e ".[] | select(.ifindex == 64)" | grep "napi-id" | wc -l
56

>From driver POV, such queries are supported by associating the pools with
an unregistered netdevice (dummy netdevice). The following limitations
are caused by such implementation:
1. The get command output specifies the 'ifindex' as 0, which is
meaningless. `iproute2` will print this as "*", but there might be other
tools which fail in such case.
2. get command does not work when devlink instance is reloaded to namespace
which is not the initial one, as the dummy device associated with the pools
belongs to the initial namespace.
See examples in commit messages.

We would like to expose page pool stats and info via the standard
interface, but such implementation is not perfect. An additional option
is to use debugfs, but we prefer to avoid it, if it is possible. Any
suggestions for better implementation in case of pool for several
netdevices will be welcomed.

Patch set overview:
Patch #1 makes netlink filling code more flex
Patch #2 changes the 'ifindex' which is used for dump
Patch #3 sets netdevice for page pools in mlxsw driver, to allow "do"
commands
Patch #4 sets page pools list for netdevices in mlxsw driver, to allow
"dump" commands

Amit Cohen (4):
  net: core: page_pool_user: Allow flexibility of 'ifindex' value
  net: core: page_pool_user: Change 'ifindex' for page pool dump
  mlxsw: pci: Allow get page pool info/stats via netlink
  mlxsw: Set page pools list for netdevices

 drivers/net/ethernet/mellanox/mlxsw/core.c    |  6 +++++
 drivers/net/ethernet/mellanox/mlxsw/core.h    |  2 ++
 drivers/net/ethernet/mellanox/mlxsw/pci.c     |  9 ++++++++
 .../net/ethernet/mellanox/mlxsw/spectrum.c    |  2 ++
 net/core/page_pool_user.c                     | 22 +++++++++----------
 5 files changed, 29 insertions(+), 12 deletions(-)

-- 
2.45.1


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ