| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
When a GSI channel is initially allocated, and after it has been
reset, the hardware assumes its ring index is 0. And although we
do initialize channels this way, the comments in the IPA code don't
really explain this. For event rings, it doesn't matter what value
we use initially, so using 0 is just fine.
Add some information about the assumptions made by hardware above
the definition of the gsi_ring structure in "gsi.h".
Zero the index field for all rings (channel and event) when the ring
is allocated. As a result, that function initializes all fields in
the structure.
Stop zeroing the index the top of gsi_channel_program(). Initially
we'll use the index value set when the channel ring was allocated.
And we'll explicitly zero the index value in gsi_channel_reset()
before programming the hardware, adding a comment explaining why
it's required.
For event rings, use the index initialized by gsi_ring_alloc()
rather than 0 when ringing the doorbell in gsi_evt_ring_program().
(It'll still be zero, but we won't assume that to be the case.)
Use a local variable in gsi_evt_ring_program() that represents the
address of the event ring's ring structure.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
mlxsw needs to distinguish nexthops with a gateway from connected
nexthops in order to write the former to the adjacency table of the
device. The check used to rely on the fact that nexthops with a gateway
have a 'link' scope whereas connected nexthops have a 'host' scope. This
is no longer correct after commit 747c14307214 ("ip: fix dflt addr
selection for connected nexthop").
Fix that by instead checking the address family of the gateway IP. This
is a more direct way and also consistent with the IPv6 counterpart in
mlxsw_sp_rt6_is_gateway().
Cc: stable@vger.kernel.org
Fixes: 747c14307214 ("ip: fix dflt addr selection for connected nexthop")
Fixes: 597cfe4fc339 ("nexthop: Add support for IPv4 nexthops")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The cited commit refactored the flow action initialization sequence to
use an interface method when translating tc action instances to flow
offload objects. The refactored version skips the initialization of the
generic flow action attributes for tc actions, such as pedit, that allocate
more than one offload entry. This can cause potential issues for drivers
mapping flow action ids.
Populate the generic flow action fields for all the flow action entries.
Fixes: c54e1d920f04 ("flow_offload: add ops to tc_action_ops for flow action setup")
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
----
v1 -> v2:
- coalese the generic flow action fields initialization to a single loop
Reviewed-by: Baowen Zheng <baowen.zheng@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For SFP port prestera driver will use kernel
phylink infrastucture to configure port mode based on
the module that has beed inserted
Co-developed-by: Yevhen Orlov <yevhen.orlov@plvision.eu>
Signed-off-by: Yevhen Orlov <yevhen.orlov@plvision.eu>
Co-developed-by: Taras Chornyi <taras.chornyi@plvision.eu>
Signed-off-by: Taras Chornyi <taras.chornyi@plvision.eu>
Signed-off-by: Oleksandr Mazur <oleksandr.mazur@plvision.eu>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Some tools (e.g. libxdp) use that information.
Signed-off-by: Andrey Turkin <andrey.turkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next
Marc Kleine-Budde says:
====================
this is a pull request of 29 patches for net-next/master.
The first 6 patches target the slcan driver. Dan Carpenter contributes
a hardening patch, followed by 5 cleanup patches.
Biju Das contributes 5 patches to prepare the sja1000 driver to
support the Renesas RZ/N1 SJA1000 CAN controller.
Dario Binacchi's patch for the slcan driver fixes a sleep with held
spin lock.
Another patch by Dario Binacchi fixes a wrong comment in the c_can
driver.
Pavel Pisa updates the CTU CAN FD IP core registers.
Stephane Grosjean contributes 3 patches to the peak_usb driver for
cleanups and support of a new MCU.
The last 12 patches are by Vincent Mailhol, they fix and improve the
txerr and rxerr reporting in all CAN drivers.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Kuniyuki Iwashima says:
====================
sysctl: Fix data-races around ipv4_net_table (Round 4).
This series fixes data-races around 17 knobs after fib_multipath_use_neigh
in ipv4_net_table.
tcp_fack was skipped because it's obsolete and there's no readers.
So, round 5 will start with tcp_dsack, 2 rounds left for 27 knobs.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_max_reordering, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its readers.
Fixes: dca145ffaa8d ("tcp: allow for bigger reordering level")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_abort_on_overflow, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_rfc1337, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_stdurg, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_retrans_collapse, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its reader.
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_slow_start_after_idle, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its readers.
Fixes: 35089bb203f4 ("[TCP]: Add tcp_slow_start_after_idle sysctl.")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_thin_linear_timeouts, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its reader.
Fixes: 36e31b0af587 ("net: TCP thin linear timeouts")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_recovery, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its readers.
Fixes: 4f41b1c58a32 ("tcp: use RACK to detect losses")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_tcp_early_retrans, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: eed530b6c676 ("tcp: early retransmit")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading these knobs, they can be changed concurrently.
Thus, we need to add READ_ONCE() to their readers.
- tcp_sack
- tcp_window_scaling
- tcp_timestamps
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_udp_l3mdev_accept, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: 63a6fff353d0 ("net: Avoid receiving packets with an l3mdev on unbound UDP sockets")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
sysctl_ip_prot_sock is accessed concurrently, and there is always a chance
of data-race. So, all readers and writers need some basic protection to
avoid load/store-tearing.
Fixes: 4548b683b781 ("Introduce a sysctl that modifies the value of PROT_SOCK.")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_fib_multipath_hash_fields, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its readers.
Fixes: ce5c9c20d364 ("ipv4: Add a sysctl to control multipath hash fields")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_fib_multipath_hash_policy, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its readers.
Fixes: bf4e0a3db97e ("net: ipv4: add support for ECMP hash policy choice")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While reading sysctl_fib_multipath_use_neigh, it can be changed
concurrently. Thus, we need to add READ_ONCE() to its reader.
Fixes: a6db4494d218 ("net: ipv4: Consider failed nexthops in multipath routes")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec
Steffen Klassert says:
====================
pull request (net): ipsec 2022-07-20
1) Fix a policy refcount imbalance in xfrm_bundle_lookup.
From Hangyu Hua.
2) Fix some clang -Wformat warnings.
Justin Stitt
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We should use of_node_put() for the reference returned by
of_get_child_by_name() which has increased the refcount.
Fixes: 45721c406dcf ("can: rcar_canfd: Add support for r8a779a0 SoC")
Link: https://lore.kernel.org/all/20220712095623.364287-1-windhl@126.com
Signed-off-by: Liang He <windhl@126.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
In commit c6f2a617a0a8 ("can: mcp251xfd: add support for mcp251863")
support for the mcp251863 was added. However it was not taken into
account that the auto detection of the chip model cannot distinguish
between mcp2518fd and mcp251863 and would lead to a warning message if
the firmware specifies a mcp251863.
Fix auto detection: If a mcp2518fd compatible chip is found, keep the
mcp251863 if specified by firmware, use mcp2518fd instead.
Link: https://lore.kernel.org/all/20220706064835.1848864-1-mkl@pengutronix.de
Fixes: c6f2a617a0a8 ("can: mcp251xfd: add support for mcp251863")
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
'can-error-set-of-fixes-and-improvement-on-txerr-and-rxerr-reporting'
Vincent Mailhol says:
====================
can: error: set of fixes and improvement on txerr and rxerr reporting
This series is a collection of patches targeting the CAN error
counter. The series is split in three blocks (with small relation to
each other).
Several drivers uses the data[6] and data[7] fields (both of type u8)
of the CAN error frame to report those values. However, the maximum
size an u8 can hold is 255 and the error counter can exceed this value
if bus-off status occurs. As such, the first nine patches of this
series make sure that no drivers try to report txerr or rxerr through
the CAN error frame when bus-off status is reached.
can_frame::data[5..7] are defined as being "controller
specific". Controller specific behaviors are not something desirable
(portability issue...) The tenth patch of this series specifies how
can_frame::data[5..7] should be use and remove any "controller
specific" freedom. The eleventh patch adds a flag to notify though
can_frame::can_id that data[6..7] were populated (in order to be
consistent with other fields).
Finally, the twelfth and last patch add three macro values to specify
the different error counter threshold with so far was hard-coded as
magic numbers in the drivers.
N.B.:
* patches 1 to 10 are for net (stable).
* patches 11 and 12 are for net-next (but depends on patches 1 to 10).
** Changelog **
v1 -> v2: https://lore.kernel.org/all/20220712153157.83847-1-mailhol.vincent@wanadoo.fr
* Fix typo in patch #10: data[7] of CAN error frames is for the RX
error counter, not the TX one (this is litteraly a one byte
change).
====================
As discussed take the whole series via can-next -> net-next.
Link: https://lore.kernel.org/all/20220719143550.3681-1-mailhol.vincent@wanadoo.fr
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
Currently, drivers are using magic numbers to derive the CAN error
states from the error counter. Add three macro declarations to
remediate this.
For reference, the error-active, error-passive and bus-off are defined
in ISO 11898, section 12.1.4.2 "Error counting". Although ISO 11898
does not define error-warning state, this extra value is also commonly
used and is thus also added.
Link: https://lore.kernel.org/all/20220719143550.3681-13-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
Add a dedicated flag in uapi/linux/can/error.h to notify the userland
that fields data[6] and data[7] of the CAN error frame were
respectively populated with the tx and rx error counters.
For all driver tree-wide, set up this flags whenever needed.
Link: https://lore.kernel.org/all/20220719143550.3681-12-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
Currently, data[5..7] of struct can_frame, when used as a CAN error
frame, are defined as being "controller specific". Device specific
behaviours are problematic because it prevents someone from writing
code which is portable between devices.
As a matter of fact, data[5] is never used, data[6] is always used to
report TX error counter and data[7] is always used to report RX error
counter. can-utils also relies on this.
This patch updates the comment in the uapi header to specify that
data[5] is reserved (and thus should not be used) and that data[6..7]
are used for error counters.
Fixes: 0d66548a10cb ("[CAN]: Add PF_CAN core module")
Link: https://lore.kernel.org/all/20220719143550.3681-11-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: 0024d8ad1639 ("can: usb_8dev: Add support for USB2CAN interface from 8 devices")
Link: https://lore.kernel.org/all/20220719143550.3681-10-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: 7259124eac7d1 ("can: kvaser_usb: Split driver into kvaser_usb_core.c and kvaser_usb_leaf.c")
Link: https://lore.kernel.org/all/20220719143550.3681-9-mailhol.vincent@wanadoo.fr
CC: Jimmy Assarsson <extja@kvaser.com>
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: aec5fb2268b7 ("can: kvaser_usb: Add support for Kvaser USB hydra family")
Link: https://lore.kernel.org/all/20220719143550.3681-8-mailhol.vincent@wanadoo.fr
CC: Jimmy Assarsson <extja@kvaser.com>
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: 0738eff14d81 ("can: Allwinner A10/A20 CAN Controller support - Kernel module")
Link: https://lore.kernel.org/all/20220719143550.3681-7-mailhol.vincent@wanadoo.fr
CC: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: 57e83fb9b746 ("can: hi311x: Add Holt HI-311x CAN driver")
Link: https://lore.kernel.org/all/20220719143550.3681-6-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
alloc_can_err_skb() already sets cf to NULL if the allocation fails [1],
so the redundant cf = NULL assignment gets removed.
[1] https://elixir.bootlin.com/linux/latest/source/drivers/net/can/dev/skb.c#L187
Fixes: 0a9cdcf098a4 ("can: slcan: extend the protocol with CAN state info")
Link: https://lore.kernel.org/all/20220719143550.3681-5-mailhol.vincent@wanadoo.fr
CC: Dario Binacchi <dario.binacchi@amarulasolutions.com>
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: 215db1856e83 ("can: sja1000: Consolidate and unify state change handling")
Link: https://lore.kernel.org/all/20220719143550.3681-4-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: fd1159318e55 ("can: add Renesas R-Car CAN driver")
Link: https://lore.kernel.org/all/20220719143550.3681-3-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
During bus off, the error count is greater than 255 and can not fit in
a u8.
Fixes: 0c78ab76a05c ("pch_can: Add setting TEC/REC statistics processing")
Link: https://lore.kernel.org/all/20220719143550.3681-2-mailhol.vincent@wanadoo.fr
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue
Tony Nguyen says:
====================
1GbE Intel Wired LAN Driver Updates 2022-07-18
This series contains updates to igc driver only.
Kurt Kanzenbach adds support for Qbv schedules where one queue stays open
in consecutive entries.
Sasha removes an unused define and field.
* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
igc: Remove forced_speed_duplex value
igc: Remove MSI-X PBA Clear register
igc: Lift TAPRIO schedule restriction
====================
Link: https://lore.kernel.org/r/20220718180109.4114540-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2022-07-18
This series contains updates to iavf driver only.
Przemyslaw fixes handling of multiple VLAN requests to account for
individual errors instead of rejecting them all. He removes incorrect
implementations of ETHTOOL_COALESCE_MAX_FRAMES and
ETHTOOL_COALESCE_MAX_FRAMES_IRQ.
He also corrects an issue with NULL pointer caused by improper handling of
dummy receive descriptors. Finally, he corrects debug prints reporting an
unknown state.
* '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
iavf: Fix missing state logs
iavf: Fix handling of dummy receive descriptors
iavf: Disallow changing rx/tx-frames and rx/tx-frames-irq
iavf: Fix VLAN_V2 addition/rejection
====================
Link: https://lore.kernel.org/r/20220718174807.4113582-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
UDP doesn't support tx memory accounting, and sysctl udp_wmem_min
is not really used anywhere. So we should fix the description in
ip-sysctl.rst accordingly.
Fixes: 95766fff6b9a ("[UDP]: Add memory accounting.")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Link: https://lore.kernel.org/r/c880a963d9b1fb5f442ae3c9e4dfa70d45296a16.1658167019.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
mtk_flow_get_wdma_info
odev pointer can be NULL in mtk_flow_offload_replace routine according
to the flower action rules. Fix possible NULL pointer dereference in
mtk_flow_get_wdma_info.
Fixes: a333215e10cb5 ("net: ethernet: mtk_eth_soc: implement flow offloading to WED devices")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/4e1685bc4976e21e364055f6bee86261f8f9ee93.1658137753.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
the last caller has been removed with commit 96f5e66e8a79 ("mac80211: fix
aggregation for hardware with ampdu queues"), so it's safe to remove this
function.
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Link: https://lore.kernel.org/r/703d549e3088367651d92a059743f1be848d74b7.1658133689.git.dcaratti@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This fixes that the platform is waked by an unexpected packet. The
size and range of FIFO is different when the device enters S3 state,
so it is necessary to correct some settings when suspending.
Regardless of jumbo frame, set RMS to 1522 and MTPS to MTPS_DEFAULT.
Besides, enable MCU_BORW_EN to update the method of calculating the
pointer of data. Then, the hardware could get the correct data.
Fixes: 195aae321c82 ("r8152: support new chips")
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Link: https://lore.kernel.org/r/20220718082120.10957-391-nic_swsd@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/kuba/linux
Pavel Begunkov says:
====================
io_uring zerocopy send
The patchset implements io_uring zerocopy send. It works with both registered
and normal buffers, mixing is allowed but not recommended. Apart from usual
request completions, just as with MSG_ZEROCOPY, io_uring separately notifies
the userspace when buffers are freed and can be reused (see API design below),
which is delivered into io_uring's Completion Queue. Those "buffer-free"
notifications are not necessarily per request, but the userspace has control
over it and should explicitly attaching a number of requests to a single
notification. The series also adds some internal optimisations when used with
registered buffers like removing page referencing.
From the kernel networking perspective there are two main changes. The first
one is passing ubuf_info into the network layer from io_uring (inside of an
in kernel struct msghdr). This allows extra optimisations, e.g. ubuf_info
caching on the io_uring side, but also helps to avoid cross-referencing
and synchronisation problems. The second part is an optional optimisation
removing page referencing for requests with registered buffers.
Benchmarking UDP with an optimised version of the selftest (see [1]), which
sends a bunch of requests, waits for completions and repeats. "+ flush" column
posts one additional "buffer-free" notification per request, and just "zc"
doesn't post buffer notifications at all.
NIC (requests / second):
IO size | non-zc | zc | zc + flush
4000 | 495134 | 606420 (+22%) | 558971 (+12%)
1500 | 551808 | 577116 (+4.5%) | 565803 (+2.5%)
1000 | 584677 | 592088 (+1.2%) | 560885 (-4%)
600 | 596292 | 598550 (+0.4%) | 555366 (-6.7%)
dummy (requests / second):
IO size | non-zc | zc | zc + flush
8000 | 1299916 | 2396600 (+84%) | 2224219 (+71%)
4000 | 1869230 | 2344146 (+25%) | 2170069 (+16%)
1200 | 2071617 | 2361960 (+14%) | 2203052 (+6%)
600 | 2106794 | 2381527 (+13%) | 2195295 (+4%)
Previously it also brought a massive performance speedup compared to the
msg_zerocopy tool (see [3]), which is probably not super interesting. There
is also an additional bunch of refcounting optimisations that was omitted from
the series for simplicity and as they don't change the picture drastically,
they will be sent as follow up, as well as flushing optimisations closing the
performance gap b/w two last columns.
For TCP on localhost (with hacks enabling localhost zerocopy) and including
additional overhead for receive:
IO size | non-zc | zc
1200 | 4174 | 4148
4096 | 7597 | 11228
Using a real NIC 1200 bytes, zc is worse than non-zc ~5-10%, maybe the
omitted optimisations will somewhat help, should look better for 4000,
but couldn't test properly because of setup problems.
Links:
liburing (benchmark + tests):
[1] https://github.com/isilence/liburing/tree/zc_v4
kernel repo:
[2] https://github.com/isilence/linux/tree/zc_v4
RFC v1:
[3] https://lore.kernel.org/io-uring/cover.1638282789.git.asml.silence@gmail.com/
RFC v2:
https://lore.kernel.org/io-uring/cover.1640029579.git.asml.silence@gmail.com/
Net patches based:
git@github.com:isilence/linux.git zc_v4-net-base
or
https://github.com/isilence/linux/tree/zc_v4-net-base
API design overview:
The series introduces an io_uring concept of notifactors. From the userspace
perspective it's an entity to which it can bind one or more requests and then
requesting to flush it. Flushing a notifier makes it impossible to attach new
requests to it, and instructs the notifier to post a completion once all
requests attached to it are completed and the kernel doesn't need the buffers
anymore.
Notifications are stored in notification slots, which should be registered as
an array in io_uring. Each slot stores only one notifier at any particular
moment. Flushing removes it from the slot and the slot automatically replaces
it with a new notifier. All operations with notifiers are done by specifying
an index of a slot it's currently in.
When registering a notification the userspace specifies a u64 tag for each
slot, which will be copied in notification completion entries as
cqe::user_data. cqe::res is 0 and cqe::flags is equal to wrap around u32
sequence number counting notifiers of a slot.
====================
Link: https://lore.kernel.org/r/cover.1657643355.git.asml.silence@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Teach tcp how to use external ubuf_info provided in msghdr and
also prepare it for managed frags by sprinkling
skb_zcopy_downgrade_managed() when it could mix managed and not managed
frags.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Teach ipv6/udp how to use external ubuf_info provided in msghdr and
also prepare it for managed frags by sprinkling
skb_zcopy_downgrade_managed() when it could mix managed and not managed
frags.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Teach ipv4/udp how to use external ubuf_info provided in msghdr and
also prepare it for managed frags by sprinkling
skb_zcopy_downgrade_managed() when it could mix managed and not managed
frags.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Managed pages contain pinned userspace pages and controlled by upper
layers, there is no need in tracking skb->pfmemalloc for them. Introduce
a helper for filling frags but ignoring page tracking, it'll be needed
later.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Some users like io_uring can do page pinning more efficiently, so we
want a way to delegate referencing to other subsystems. For that add
a new flag called SKBFL_MANAGED_FRAG_REFS. When set, skb doesn't hold
page references and upper layers are responsivle to managing page
lifetime.
It's allowed to convert skbs from managed to normal by calling
skb_zcopy_downgrade_managed(). The function will take all needed
page references and clear the flag. It's needed, for instance,
to avoid mixing managed modes.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
