When no H2G transport is loaded, vsock currently routes all CIDs to the
G2H transport (commit 65b422d9b6 ("vsock: forward all packets to the
host when no H2G is registered"). Extend that existing behavior: when
an H2G transport is loaded but does not claim a given CID, the
connection falls back to G2H in the same way.
This matters in environments like Nitro Enclaves, where an instance may
run nested VMs via vhost-vsock (H2G) while also needing to reach sibling
enclaves at higher CIDs through virtio-vsock-pci (G2H). With the old
code, any CID > 2 was unconditionally routed to H2G when vhost was
loaded, making those enclaves unreachable without setting
VMADDR_FLAG_TO_HOST explicitly on every connect.
Requiring every application to set VMADDR_FLAG_TO_HOST creates friction:
tools like socat, iperf, and others would all need to learn about it.
The flag was introduced 6 years ago and I am still not aware of any tool
that supports it. Even if there was support, it would be cumbersome to
use. The most natural experience is a single CID address space where H2G
only wins for CIDs it actually owns, and everything else falls through to
G2H, extending the behavior that already exists when H2G is absent.
To give user space at least a hint that the kernel applied this logic,
automatically set the VMADDR_FLAG_TO_HOST on the remote address so it
can determine the path taken via getpeername().
Add a per-network namespace sysctl net.vsock.g2h_fallback (default 1).
At 0 it forces strict routing: H2G always wins for CID > VMADDR_CID_HOST,
or ENODEV if H2G is not loaded.
Signed-off-by: Alexander Graf <graf@amazon.com>
Tested-by: syzbot@syzkaller.appspotmail.com
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://patch.msgid.link/20260304230027.59857-1-graf@amazon.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Commit dd23c9f1e8 ("tcp: annotate data-races around tp->tsoffset")
updated do_tcp_getsockopt() to read tp->tsoffset with READ_ONCE()
for TCP_TIMESTAMP because another CPU may change it concurrently.
tcp_v6_connect() still stores tp->tsoffset with a plain write. That
store runs under lock_sock() via inet_stream_connect(), but the socket
lock does not serialize a concurrent getsockopt(TCP_TIMESTAMP) from
another task sharing the socket.
Use WRITE_ONCE() for the tcp_v6_connect() store so the connect-time
writer matches the lockless TCP_TIMESTAMP reader. This also makes the
IPv6 path consistent with tcp_v4_connect().
Signed-off-by: Wesley Atwell <atwellwea@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jiayuan Chen <jiayuan.chen@shopee.com>
Link: https://patch.msgid.link/20260310012604.145661-1-atwellwea@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Gal Pressman says:
====================
selftests: net: fix cmd.process() timeout handling
Pass the timeout argument correctly in cmd.process().
As Jakub noted, fixing the timeout broke the bpftrace() command
in netpoll_basic.py, so fix it first.
====================
Link: https://patch.msgid.link/20260310115803.2521050-1-gal@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit refactored the hardcoded timeout=5 into a parameter,
but dropped the keyword from the communicate() call.
Since Popen.communicate()'s first positional argument is 'input' (not
'timeout'), the timeout value is silently treated as stdin input and the
call never enforces a timeout.
Pass timeout as a keyword argument to restore the intended behavior.
Reviewed-by: Nimrod Oren <noren@nvidia.com>
Signed-off-by: Gal Pressman <gal@nvidia.com>
Link: https://patch.msgid.link/20260310115803.2521050-3-gal@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The bpftrace() helper configures an interval based exit timer but does
not propagate the timeout to the cmd object, which defaults to 5
seconds. Since the default BPFTRACE_TIMEOUT is 10 seconds, cmd.process()
always raises a TimeoutExpired exception before bpftrace has a chance to
exit gracefully.
Pass timeout+5 to cmd() to allow bpftrace to complete gracefully.
Note: this issue is masked by a bug in the way cmd() passes timeout,
this is fixed in the next commit.
Reviewed-by: Nimrod Oren <noren@nvidia.com>
Signed-off-by: Gal Pressman <gal@nvidia.com>
Link: https://patch.msgid.link/20260310115803.2521050-2-gal@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kevin Hao says:
====================
net: macb: Clean up several member settings of macb_config instances
While debugging an issue in the macb driver, I noticed that many macb_config
instances have very similar member settings. This makes it difficult to
identify the actual differences between these instances. This patch series
aims to clean up some of these settings and clarify the specific configurations
of each macb_config instance. No functional changes are introduced.
====================
Link: https://patch.msgid.link/20260310-macb-cleanup-v1-0-928c1a91a7dc@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All instances of macb_config currently have the .usrio set, but most of
them use &macb_default_usrio. In fact, there is no need to duplicate
this across all macb_config instances. Remove the .usrio setting from
instances that use &macb_default_usrio, and ensure that the default is
selected at runtime when no other value is explicitly set.
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Link: https://patch.msgid.link/20260310-macb-cleanup-v1-3-928c1a91a7dc@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All instances of macb_config currently have the .init field set, but most
of them use macb_init(). In fact, there is no need to duplicate this
across all macb_config instances. Introduce a new macb_init() function
that executes the specific .init if it is set; otherwise, it runs a
default initialization function.
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Link: https://patch.msgid.link/20260310-macb-cleanup-v1-2-928c1a91a7dc@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All instances of macb_config currently have .clk_init set, but most of
them use macb_clk_init(). In fact, there is no need to duplicate this
across all macb_config instances. Introduce a new macb_clk_init()
function that executes the specific .clk_init if it is set; otherwise,
it runs the default clock initialization function.
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Link: https://patch.msgid.link/20260310-macb-cleanup-v1-1-928c1a91a7dc@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add tests to local_termination.sh to verify that link-local frames
arrive. On some switches the DSA driver uses bridges to connect the
user ports to their CPU ports. More "intelligent" switches typically
don't forward link-local frames, but may trap them to an internal
microcontroller. The driver may have to change trapping rules, so
link-local frames end up on the DSA CPU ports instead of being
silently dropped or trapped to the internal microcontroller of the
switch.
Add two tests which help to validate this has been done correctly:
- Link-local STP BPDU should arrive at the Linux netdev when the
bridge has STP disabled (BR_NO_STP), in which case the bridge
forwards them rather than consuming them in the control plane
- Link-local LLDP should arrive at standalone ports (and the test
should be skipped on bridged ports similar to how it is done
for the IEEE1588v2/PTP tests)
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/1a67081b2ede1e6d2d32f7dd54ae9688f3566152.1773166131.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Russell King says:
====================
net: stmmac: start to shrink memory usage
Start shrinking stmmac's memory usage by avoiding using "int" for
members that are only used for 0/1 (boolean) values, or values that
can't be larger than 255.
In addition, as struct stmmac_dma_cfg is approximately a cache line,
shrinks below a cache line as a result of this patch set, and is
required, there is no point separately allocating this from
struct plat_stmmacenet_data. Embed it into the end of this struct
and set the existing pointer to avoid large wide-spread changes.
Lastly, add documentation for struct stmmac_dma_cfg, and document
the stmmac clocks as best we can given the driver history.
====================
Link: https://patch.msgid.link/aa6VEsmBK-S9eNYU@shell.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add documentation covering stmmac_clk, pclk, clk_ptp_ref and clk_tx_i
in the hope that this will help understand what each of these clocks
are for.
There is confusion around stmmac_clk and pclk which can't be easily
resolved today as the Imagination Technologies Pistachio board that
pclk was introduced for has no public documentation and is likely now
obsolete. So the origins of pclk are lost to the winds of time.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Tested-by: Mohd Ayaan Anwar <mohd.anwar@oss.qualcomm.com>
Link: https://patch.msgid.link/E1vzX5Z-0000000CVsb-1XTm@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The maximum number of queues is a compile time constant of only eight.
This makes using a 32-bit quantity wastefulf. Instead, use u8 for
these and their associated variables.
When reading the DT properties, saturdate at U8_MAX. Provided the core
provides DMA capabilities to describe the number of queues, this will
be capped by stmmac_hw_init() with a warning.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Tested-by: Mohd Ayaan Anwar <mohd.anwar@oss.qualcomm.com>
Link: https://patch.msgid.link/E1vzX5K-0000000CVsE-0J0Y@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
plat_dat->dma_cfg is unconditionally required for the operation of the
driver, so it would make sense to allocate it along with the plat_dat.
On Arm64, sizeof(*plat_dat) has recently shrunk from 880 to 816 bytes
and sizeof(*plat_dat->dma_cfg) has shrunk from 32 to 20 bytes.
Given that dma_cfg is required, and it is now less than a cache line,
It doesn't make sense to allocate this separateny, so place it at the
end of struct plat_stmmacenet_data, and set plat_dat->dma_cfg to point
at that to avoid mass changes.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Mohd Ayaan Anwar <mohd.anwar@oss.qualcomm.com>
Tested-by: Mohd Ayaan Anwar <mohd.anwar@oss.qualcomm.com>
Link: https://patch.msgid.link/E1vzX54-0000000CVrw-2jfu@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Driver core holds a reference to the USB interface and its parent USB
device while the interface is bound to a driver and there is no need to
take additional references unless the structures are needed after
disconnect.
Drop the redundant device reference to reduce cargo culting, make it
easier to spot drivers where an extra reference is needed, and reduce
the risk of memory leaks when drivers fail to release it.
Signed-off-by: Johan Hovold <johan@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260309082641.15574-1-johan@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Raju Rangoju says:
====================
amd-xgbe: Improve power management for S0i3
Improve the amd-xgbe power management handling to allow AMD platforms to
reach the deepest suspend state (S0i3) when modern standby is used.
The first patch cleans up the xgbe_powerdown() and xgbe_powerup()
helpers by removing an unused caller distinction and aligning the
ordering of operations with xgbe_stop().
The second patch adds proper PCI power management operations, following
the standard PCI PM model, so that the device can be cleanly put into
D3 and resumed back to D0. Without this, the amd_pmc driver reports:
"Last suspend didn't reach deepest state"
when the amd-xgbe driver is enabled.
These changes have been tested on AMD platforms using S0i3 modern
standby.
====================
Link: https://patch.msgid.link/20260308092851.1510214-1-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The current suspend/resume implementation does not correctly handle PCI
device power state transitions, which prevents AMD platforms from
reaching the deepest suspend state (S0i3) when the amd-xgbe driver is
enabled.
In particular, the amd_pmc driver reports:
"Last suspend didn't reach deepest state"
when this device is present.
Implement proper PCI power management operations following the standard
PCI PM model so that the device can be cleanly powered down and resumed.
Suspend path:
- Power down the network interface
- Put the PHY into low-power mode
- Disable bus mastering to prevent DMA activity
- Save PCI configuration space
- Disable the PCI device
- Disable wake from D3 (S0i3 does not require Wake-on-LAN)
- Set the device to D3hot
Resume path:
- Restore the PCI power state to D0
- Restore PCI configuration space
- Enable the PCI device
- Re-enable bus mastering
- Re-enable device interrupts
- Clear the PHY low-power mode
- Power up the network interface
This allows systems using amd-xgbe to reach the deepest suspend state
when entering modern standby (S0i3).
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260308092851.1510214-3-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The caller parameter in xgbe_powerdown() and xgbe_powerup() was intended
to differentiate between driver and ioctl contexts, but the only
remaining usage is from the driver suspend/resume path.
Simplify this by:
- Removing the unused XGMAC_DRIVER_CONTEXT and XGMAC_IOCTL_CONTEXT
macros
- Dropping the now-unused caller parameter
- Reordering operations in xgbe_powerdown() to disable NAPI before
stopping TX/RX, matching the order used in xgbe_stop()
This makes the powerdown/powerup paths easier to follow and keeps the
ordering consistent with the rest of the driver.
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260308092851.1510214-2-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The device firmware supports reporting more media types than what was
there in the past, so map these new media types to existing ethtool
bits, which appears to be what other drivers do for media types that
match speeds but not physical spec.
And while here, make a very small cleanup in ionic_get_link_ksettings()
to remove some unnecessary code duplication.
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
Link: https://patch.msgid.link/20260306215634.64550-1-eric.joyner@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Jakub Kicinski says:
====================
tools: ynl: policy query support
Improve the Netlink policy support in YNL. This series grew out of
improvements to policy checking, when writing selftests I realized
that instead of doing all the policy parsing in the test we're
better off making it part of YNL itself.
Patch 1 adds pad handling, apparently we never hit pad with commonly
used families. nlctrl policy dumps use pad more frequently.
Patch 2 is a trivial refactor.
Patch 3 pays off some technical debt in terms of documentation.
The YnlFamily class is growing in size and it's quite hard to
find its members. So document it a little bit.
Patch 4 is the main dish, the implementation of get_policy(op)
in YnlFamily.
Patch 5 plugs the new functionality into the CLI.
====================
Link: https://patch.msgid.link/20260310005337.3594225-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The format of Netlink policy dump is a bit curious with messages
in the same dump carrying both attrs and mapping info. Plus each
message carries a single piece of the puzzle the caller must then
reassemble.
I need to do this reassembly for a test, but I think it's generally
useful. So let's add proper support to YnlFamily to return more
user-friendly representation. See the various docs in the patch
for more details.
Link: https://patch.msgid.link/20260310005337.3594225-5-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If nvmem loads after the ethernet driver, mac address assignments will
not take effect. of_get_ethdev_address returns EPROBE_DEFER in such a
case so we need to handle that to avoid eth_hw_addr_random.
Add extra goto section to just free stats as they are allocated right
above.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260307031709.640141-1-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The test generates 16 flows, and verifies that traffic is distributed
across two queues via the NICs RSS indirection table. The likelihood of the
flows skewing to a single queue is high, so we retry sending traffic up to
3 times.
Alternatively, we could increase the number of generated flows. But
debug kernels may struggle to ramp this many flows.
During manual testing, the test passed for 10,000 consecutive runs.
Signed-off-by: Dimitri Daskalakis <dimitri.daskalakis1@gmail.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://patch.msgid.link/20260309204215.2110486-1-dimitri.daskalakis1@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With the current port selection algorithm, ports after a reserved port
range or long time used port are used more often than others [1]. This
causes an uneven port usage distribution. This combines with cloud
environments blocking connections between the application server and the
database server if there was a previous connection with the same source
port, leading to connectivity problems between applications on cloud
environments.
The real issue here is that these firewalls cannot cope with
standards-compliant port reuse. This is a workaround for such situations
and an improvement on the distribution of ports selected.
The proposed solution is to implement a variant of RFC 6056 Algorithm 5.
The step size is selected randomly on every connect() call ensuring it
is a coprime with respect to the size of the range of ports we want to
scan. This way, we can ensure that all ports within the range are
scanned before returning an error. To enable this algorithm, the user
must configure the new sysctl option "net.ipv4.ip_local_port_step_width".
In addition, on graphs generated we can observe that the distribution of
source ports is more even with the proposed approach. [2]
[1] https://0xffsoftware.com/port_graph_current_alg.html
[2] https://0xffsoftware.com/port_graph_random_step_alg.html
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Fernando Fernandez Mancera <fmancera@suse.de>
Link: https://patch.msgid.link/20260309023946.5473-2-fmancera@suse.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allison Henderson says:
====================
selftests: rds: ksft cleanups
This set addresses a few rds selftests clean ups and bugs encountered
when running in the ksft framework. The first patch is a clean up
patch that addresses pylint warnings, but otherwise no functional
changes. The next patch moves the test time out to a ksft settings
file so that the time out is set appropriately. And lastly we fix a
tcpdump segfault caused by deprecated a os.fork() call.
====================
Link: https://patch.msgid.link/20260308055835.1338257-1-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/rds/test.py sees a segfault in tcpdump when executed through the
ksft runner.
[ 21.903713] tcpdump[1469]: segfault at 0 ip 000072100e99126d
sp 00007ffccf740fd0 error 4
[ 21.903721] in libc.so.6[16a26d,7798b149a000+188000]
[ 21.905074] in libc.so.6[16a26d,72100e84f000+188000] likely on
CPU 5 (core 5, socket 0)
[ 21.905084] Code: 00 0f 85 a0 00 00 00 48 83 c4 38 89 d8 5b 41 5c
41 5d 41 5e 41 5f 5d c3 0f 1f 44 00 00 48 8b 05 91 8b 09 00 8b 4d ac
64 89 08 <41> 0f b6 07 83 e8 2b a8 fd 0f 84 54 ff ff ff 49 8b 36 4c 89
ff e8
[ 21.906760] likely on CPU 9 (core 9, socket 0)
[ 21.913469] Code: 00 0f 85 a0 00 00 00 48 83 c4 38 89 d8 5b 41 5c 41
5d 41 5e 41 5f 5d c3 0f 1f 44 00 00 48 8b 05 91 8b 09 00 8b 4d ac 64 89
08 <41> 0f b6 07 83 e8 2b a8 fd 0f 84 54 ff ff ff 49 8b 36 4c 89 ff e8
The os.fork() call creates extra complexity because it forks the entire
process including the python interpreter. ip() then calls cmd() which
creates a subprocess.Popen. We can avoid the extra layering by simply
calling subprocess.Popen directly. Track the process handles directly
and terminate them at cleanup rather than relying on killall. Further
tcpdump's -Z flag attempts to change savefile ownership, which is not
supported by the 9p protocol. Fix this by writing pcap captures to
"/tmp" during the test and move them to the log directory after tcpdump
exits.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260308055835.1338257-4-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
NIPA tries to make sure that HW tests don't modify system state.
It dumps some well known configs before and after the test and
compares the outputs.
Make sure that YNL json output is stable. Converting sets to lists
with a naive list(o) results in a random order.
Link: https://patch.msgid.link/20260307175916.1652518-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kyoji Ogasawara says:
====================
smc-sysctl formatting and missing entries
update SMC sysctl documentation in two small steps.
- patch 1 fixes indentation in the smcr_buf_type section
- patch 2 documents missing sysctl parameters limit_smc_hs and hs_ctrl,
including values/defaults and hs_ctrl usage notes
====================
Link: https://patch.msgid.link/20260309124541.22723-1-sawara04.o@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Mike Marciniszyn says:
====================
eth fbnic: Add fbnic self tests
From: "Mike Marciniszyn (Meta)" <mike.marciniszyn@gmail.com>
This series adds self tests to test the registers, the
msix interrupts, the tlv, and the firmware mailbox.
This series assumes that the
[PATCH net-next 0/2] Add debugfs hooks [1]
is present.
When the self tests are run the with ethtool -t:
ethtool -t eth0
The test result is PASS
The test extra info:
Register test (offline) 0
MSI-X Interrupt test (offline) 0
FW mailbox test (on/offline) 0
====================
Link: https://patch.msgid.link/20260307105847.1438-1-mike.marciniszyn@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The register test will be used to verify hardware is behaving as expected.
The test itself will have us writing to registers that should have no
side effects due to us resetting after the test has been completed.
While the test is being run the interface should be offline.
This patch counts on the first patch of this series to export netif_open()
and also ensures that the half close calls netif_close() to
avoid deadlock.
Signed-off-by: Mike Marciniszyn (Meta) <mike.marciniszyn@gmail.com>
Link: https://patch.msgid.link/20260307105847.1438-3-mike.marciniszyn@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
