| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
This completes the previous patches by moving notifier registration for
SF dev tables outside the devlink locked critical section in
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.
This is only done for non-SFs, since SFs do not have a SF HW table
themselves.
After this patch, notifiers can grab the PF devlink lock (soon to be
necessary) without creating a locking cycle.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763325940-1231508-7-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Move the SF table notifiers registration/unregistration outside of
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.
This is only done for non-SFs, since SFs do not have a SF table
themselves and thus don't need notifiers.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763325940-1231508-6-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Move the SF HW table notifier registration/unregistration outside of
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.
This is only done for non-SFs, since SFs do not have a SF HW table
themselves.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763325940-1231508-5-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The vhca event notifier consists of an atomic notifier for vhca state
changes (used for SF events), multiple workqueues and a blocking
notifier chain for delivering the vhca state change events for further
processing.
This patch moves the vhca notifier head outside of mlx5_init_one() /
mlx5_uninit_one() and into the mlx5_mdev_init() / mlx5_mdev_uninit()
functions.
This allows called notifiers to grab the PF devlink lock which was
previously impossible because it would create a circular lock
dependency.
mlx5_vhca_event_stop() is now called earlier in the cleanup phase and
flushes the workqueues to ensure that after the call, there are no
pending events. This simplifies the cleanup flow for vhca event
consumers.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763325940-1231508-4-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The esw mode change notifier chain is initialized/cleaned up in
mlx5_init_one() / mlx5_uninit_one() with the devlink lock held.
Move the notifier head from the eswitch struct into mlx5_priv directly,
and initialize it outside the critical section. This will allow notifier
registration to happen earlier in the init procedure in subsequent
patches.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763325940-1231508-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Move event init/cleanup outside of mlx5_init_one() / mlx5_uninit_one()
and into the mlx5_mdev_init() / mlx5_mdev_uninit() functions.
By doing this, we avoid the events being reinitialized on devlink reload
and, more importantly, the events->sw_nh notifier chain becomes
available earlier in the init procedure, which will be used in
subsequent patches. This makes sense because the events struct is pure
software, independent of any HW details.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763325940-1231508-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Jason Xing says:
====================
net: adjust conservative values around napi
This series keeps at least 96 skbs per cpu and frees 32 skbs at one
time in conclusion. More initial discussions with Eric can be seen at
the link [1].
[1]: https://lore.kernel.org/all/CAL+tcoBEEjO=-yvE7ZJ4sB2smVBzUht1gJN85CenJhOKV====================
Link: https://patch.msgid.link/20251118070646.61344-1-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
After getting the current skb in napi_skb_cache_get(), the next skb in
cache is highly likely to be used soon, so prefetch would be helpful.
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://patch.msgid.link/20251118070646.61344-5-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
- Replace NAPI_SKB_CACHE_HALF with NAPI_SKB_CACHE_FREE
- Only free 32 skbs in napi_skb_cache_put()
Since the first patch adjusting NAPI_SKB_CACHE_SIZE to 128, the number
of packets to be freed in the softirq was increased from 32 to 64.
Considering a subsequent net_rx_action() calling napi_poll() a few
times can easily consume the 64 available slots and we can afford
keeping a higher value of sk_buffs in per-cpu storage, decrease
NAPI_SKB_CACHE_FREE to 32 like before. So now the logic is 1) keeping
96 skbs, 2) freeing 32 skbs at one time.
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://patch.msgid.link/20251118070646.61344-4-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The previous value 16 is a bit conservative, so adjust it along with
NAPI_SKB_CACHE_SIZE, which can minimize triggering memory allocation
in napi_skb_cache_get*().
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://patch.msgid.link/20251118070646.61344-3-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
After commit b61785852ed0 ("net: increase skb_defer_max default to 128")
changed the value sysctl_skb_defer_max to avoid many calls to
kick_defer_list_purge(), the same situation can be applied to
NAPI_SKB_CACHE_SIZE that was proposed in 2016. It's a trade-off between
using pre-allocated memory in skb_cache and saving more a bit heavy
function calls in the softirq context.
With this patch applied, we can have more skbs per-cpu to accelerate the
sending path that needs to acquire new skbs.
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://patch.msgid.link/20251118070646.61344-2-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Vladimir Oltean says:
====================
Disable CLKOUT on RTL8211F(D)(I)-VD-CG
The Realtek RTL8211F(D)(I)-VD-CG is similar to other RTL8211F models in
that the CLKOUT signal can be turned off - a feature requested to reduce
EMI, and implemented via "realtek,clkout-disable" as documented in
Documentation/devicetree/bindings/net/realtek,rtl82xx.yaml.
It is also dissimilar to said PHY models because it has no PHYCR2
register, and disabling CLKOUT is done through some other register.
The strategy adopted in this 6-patch series is to make the PHY driver
not think in terms of "priv->has_phycr2" and "priv->phycr2", but of more
high-level features ("priv->disable_clk_out") while maintaining behaviour.
Then, the logic is extended for the new PHY.
Very loosely based on previous work from Clark Wang, who took a
different approach, to pretend that the RTL8211FVD_CLKOUT_REG is
actually this PHY's PHYCR2.
====================
Link: https://patch.msgid.link/20251117234033.345679-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
To simplify the rtl8211f_config_init() control flow and get rid of
"early" returns for PHYs where the PHYCR2 register is absent, move the
entire logic sub-block that deals with disabling PHY-mode EEE to a
separate function. There, it is much more obvious what the early
"return 0" skips, and it becomes more difficult to accidentally skip
unintended stuff.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251117234033.345679-7-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Previous changes have replaced the machine-level priv->phycr2 with a
high-level priv->disable_clk_out. This created a discrepancy with
priv->phycr1 which is resolved here, for uniformity.
One advantage of this new implementation is that we don't read
priv->phycr1 in rtl821x_probe() if we're never going to modify it.
We never test the positive return code from phy_modify_mmd_changed(), so
we could just as well use phy_modify_mmd().
I took the ALDPS feature description from commit d90db36a9e74 ("net:
phy: realtek: add dt property to enable ALDPS mode") and transformed it
into a function comment - the feature is sufficiently non-obvious to
deserve that.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251117234033.345679-6-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add CLKOUT disable support for RTL8211F(D)(I)-VD-CG. Like with other PHY
variants, this feature might be requested by customers when the clock
output is not used, in order to reduce electromagnetic interference (EMI).
In the common driver, the CLKOUT configuration is done through PHYCR2.
The RTL_8211FVD_PHYID is singled out as not having that register, and
execution in rtl8211f_config_init() returns early after commit
2c67301584f2 ("net: phy: realtek: Avoid PHYCR2 access if PHYCR2 not
present").
But actually CLKOUT is configured through a different register for this
PHY. Instead of pretending this is PHYCR2 (which it is not), just add
some code for modifying this register inside the rtl8211f_disable_clk_out()
function, and move that outside the code portion that runs only if
PHYCR2 exists.
In practice this reorders the PHYCR2 writes to disable PHY-mode EEE and
to disable the CLKOUT for the normal RTL8211F variants, but this should
be perfectly fine.
It was not noted that RTL8211F(D)(I)-VD-CG would need a genphy_soft_reset()
call after disabling the CLKOUT. Despite that, we do it out of caution
and for symmetry with the other RTL8211F models.
Co-developed-by: Clark Wang <xiaoning.wang@nxp.com>
Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251117234033.345679-5-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This variable is assigned in rtl821x_probe() and used in
rtl8211f_config_init(), which is more complex than it needs to be.
Simply testing the same condition from rtl821x_probe() in
rtl8211f_config_init() yields the same result (the PHY driver ID is a
runtime invariant), but with one temporary variable less.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251117234033.345679-4-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The RTL8211F(D)(I)-VD-CG PHY also has support for disabling the CLKOUT,
and we'd like to introduce the "realtek,clkout-disable" property for
that.
But it isn't done through the PHYCR2 register, and it becomes awkward to
have the driver pretend that it is. So just replace the machine-level
"u16 phycr2" variable with a logical "bool disable_clk_out", which
scales better to the other PHY as well.
The change is a complete functional equivalent. Before, if the device
tree property was absent, priv->phycr2 would contain the RTL8211F_CLKOUT_EN
bit as read from hardware. Now, we don't save priv->phycr2, but we just
don't call phy_modify_paged() on it. Also, we can simply call
phy_modify_paged() with the "set" argument to 0.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251117234033.345679-3-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The control flow in rtl8211f_config_init() has some pitfalls which were
probably unintended. Specifically it has an early return:
switch (phydev->interface) {
...
default: /* the rest of the modes imply leaving delay as is. */
return 0;
}
which exits the entire config_init() function. This means it also skips
doing things such as disabling CLKOUT or disabling PHY-mode EEE.
For the RTL8211FS, which uses PHY_INTERFACE_MODE_SGMII, this might be a
problem. However, I don't know that it is, so there is no Fixes: tag.
The issue was observed through code inspection.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251117234033.345679-2-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Convert the vmxnet3 driver to use the new .get_rx_ring_count ethtool
operation instead of implementing .get_rxnfc solely for handling
ETHTOOL_GRXRINGS command. This simplifies the code by removing the
switch statement and replacing it with a direct return of the queue
count.
The new callback provides the same functionality in a more direct way,
following the ongoing ethtool API modernization.
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20251118-vmxnet3_grxrings-v1-1-ed8abddd2d52@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Aditya Garg says:
====================
net: mana: Enforce TX SGE limit and fix error cleanup
Add pre-transmission checks to block SKBs that exceed the hardware's SGE
limit. Force software segmentation for GSO traffic and linearize non-GSO
packets as needed.
Update TX error handling to drop failed SKBs and unmap resources
immediately.
====================
Link: https://patch.msgid.link/1763464269-10431-1-git-send-email-gargaditya@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Drop TX packets when posting the work request fails and ensure DMA
mappings are always cleaned up.
Signed-off-by: Aditya Garg <gargaditya@linux.microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Link: https://patch.msgid.link/1763464269-10431-3-git-send-email-gargaditya@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The MANA hardware supports a maximum of 30 scatter-gather entries (SGEs)
per TX WQE. Exceeding this limit can cause TX failures.
Add ndo_features_check() callback to validate SKB layout before
transmission. For GSO SKBs that would exceed the hardware SGE limit, clear
NETIF_F_GSO_MASK to enforce software segmentation in the stack.
Add a fallback in mana_start_xmit() to linearize non-GSO SKBs that still
exceed the SGE limit.
Also, Add ethtool counter for SKBs linearized
Co-developed-by: Dipayaan Roy <dipayanroy@linux.microsoft.com>
Signed-off-by: Dipayaan Roy <dipayanroy@linux.microsoft.com>
Signed-off-by: Aditya Garg <gargaditya@linux.microsoft.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Link: https://patch.msgid.link/1763464269-10431-2-git-send-email-gargaditya@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Currently, the TM tree is printing all SQ topology including those
which are not enabled, this results in redundant output for SQs
which are not active. This patch adds a check in print_tm_tree()
to skip printing the TM tree hierarchy if the SQ is not enabled.
Signed-off-by: Anshumali Gaur <agaur@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251118054235.1599714-1-agaur@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Different SoCs have different numbers of Wireless Ethernet
Dispatch (WED) units:
- MT7981: Has 1 WED unit
- MT7986: Has 2 WED units
- MT7988: Has 2 WED units
Update the binding to reflect these hardware differences. The MT7981
also uses infracfg for PHY switching, so allow that property.
Signed-off-by: Sjoerd Simons <sjoerd@collabora.com>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://patch.msgid.link/20251115-openwrt-one-network-v4-6-48cbda2969ac@collabora.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Russell King says:
====================
net: stmmac: sanitise stmmac_is_jumbo_frm()
stmmac_is_jumbo_frm() takes skb->len, which is unsigned int, but the
parameter is passed as an "int" and then tested using signed
comparisons. This can cause bugs. Change the parameter to be unsigned.
Also arrange for it to return a bool.
====================
Link: https://patch.msgid.link/aRxDqJSWxOdOaRt4@shell.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
stmmac_is_jumbo_frm() returns whether the driver considers the frame
size to be a jumbo frame, and thus returns 0/1 values. This is boolean,
so convert it to return a boolean and use false/true instead. Also
convert stmmac_xmit()'s is_jumbo to be bool, which causes several
variables to be repositioned to keep it in reverse Christmas-tree
order.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/E1vLIWW-0000000Ewkl-21Ia@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
stmmac_is_jumbo_frm() and the is_jumbo_frm() methods take skb->len
which is an unsigned int. Avoid an implicit cast to "int" via the
method parameter and then incorrectly doing signed comparisons on
this unsigned value.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/E1vLIWR-0000000Ewkf-1Tdx@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
priv->sph* only have 'true' and 'false' used with them, yet they are an
int. Change their type to a bool, and rename to make their usage more
clear.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/E1vLIDN-0000000Evur-2NLU@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The new test checks that a route that has been promoted from RA-learned
to static does not switch back when a new RA message arrives. In
addition, it checks that the route is owned by RA again when the static
address is removed.
Signed-off-by: Fernando Fernandez Mancera <fmancera@suse.de>
Link: https://patch.msgid.link/20251115095939.6967-2-fmancera@suse.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
When an IPv6 Router Advertisement (RA) is received for a prefix, the
kernel creates the corresponding on-link route with flags RTF_ADDRCONF
and RTF_PREFIX_RT configured and RTF_EXPIRES if lifetime is set.
If later a user configures a static IPv6 address on the same prefix the
kernel clears the RTF_EXPIRES flag but it doesn't clear the RTF_ADDRCONF
and RTF_PREFIX_RT. When the next RA for that prefix is received, the
kernel sees the route as RA-learned and wrongly configures back the
lifetime. This is problematic because if the route expires, the static
address won't have the corresponding on-link route.
This fix clears the RTF_ADDRCONF and RTF_PREFIX_RT flags preventing that
the lifetime is configured when the next RA arrives. If the static
address is deleted, the route becomes RA-learned again.
Fixes: 14ef37b6d00e ("ipv6: fix route lookup in addrconf_prefix_rcv()")
Reported-by: Garri Djavadyan <g.djavadyan@gmail.com>
Closes: https://lore.kernel.org/netdev/ba807d39aca5b4dcf395cc11dca61a130a52cfd3.camel@gmail.com/
Signed-off-by: Fernando Fernandez Mancera <fmancera@suse.de>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20251115095939.6967-1-fmancera@suse.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Kuniyuki Iwashima says:
====================
af_unix: GC cleanup and optimisation.
Currently, AF_UNIX GC is triggered from close() and sendmsg()
based on the number of inflight AF_UNIX sockets.
This is because the old GC implementation had no idea of the
shape of the graph formed by SCM_RIGHTS references.
The new GC knows whether cyclic references (could) exist.
This series refines such conditions not to trigger GC unless
really needed.
====================
Link: https://patch.msgid.link/20251115020935.2643121-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
unix_schedule_gc() and wait_for_unix_gc() share some code.
Let's consolidate the two.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-8-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
unix_tot_inflight is no longer used.
Let's remove it.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-7-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
unix_tot_inflight is a poor metric, only telling the number of
inflight AF_UNXI sockets, and we should use unix_graph_state instead.
Also, if the receiver is catching up with the passed fds, the
sender does not need to schedule GC.
GC only helps unreferenced cyclic SCM_RIGHTS references, and in
such a situation, the malicious sendmsg() will continue to call
wait_for_unix_gc() and hit the UNIX_INFLIGHT_SANE_USER condition.
Let's make only malicious users schedule GC and wait for it to
finish if a cyclic reference exists during the previous GC run.
Then, sane users will pay almost no cost for wait_for_unix_gc().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-6-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
We have been calling wait_for_unix_gc() on every sendmsg() in case
there are too many inflight AF_UNIX sockets.
This is also because the old GC implementation had poor knowledge
of the inflight sockets and had to suspect every sendmsg().
This was improved by commit d9f21b361333 ("af_unix: Try to run GC
async."), but we do not even need to call wait_for_unix_gc() if the
process is not sending AF_UNIX sockets.
The wait_for_unix_gc() call only helps when a malicious process
continues to create cyclic references, and we can detect that
in a better place and slow it down.
Let's move wait_for_unix_gc() to unix_prepare_fpl() that is called
only when AF_UNIX socket fd is passed via SCM_RIGHTS.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-5-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
We have been triggering GC on every close() if there is even one
inflight AF_UNIX socket.
This is because the old GC implementation had no idea of the graph
shape formed by SCM_RIGHTS references.
The new GC knows whether there could be a cyclic reference or not,
and we can do better.
Let's not trigger GC from close() if there is no cyclic reference
or GC is already in progress.
While at it, unix_gc() is renamed to unix_schedule_gc() as it does
not actually perform GC since commit 8b90a9f819dc ("af_unix: Run
GC on only one CPU.").
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-4-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
GC manages its state by two variables, unix_graph_maybe_cyclic
and unix_graph_grouped, both of which are set to false in the
initial state.
When an AF_UNIX socket is passed to an in-flight AF_UNIX socket,
unix_update_graph() sets unix_graph_maybe_cyclic to true and
unix_graph_grouped to false, making the next GC invocation call
unix_walk_scc() to group SCCs.
Once unix_walk_scc() finishes, sockets in the same SCC are linked
via vertex->scc_entry. Then, unix_graph_grouped is set to true
so that the following GC invocations can skip Tarjan's algorithm
and simply iterate through the list in unix_walk_scc_fast().
In addition, if we know there is at least one cyclic reference,
we set unix_graph_maybe_cyclic to true so that we do not skip GC.
So the state transitions as follows:
(unix_graph_maybe_cyclic, unix_graph_grouped)
=
(false, false) -> (true, false) -> (true, true) or (false, true)
^.______________/________________/
There is no transition to the initial state where both variables
are false.
If we consider the initial state as grouped, we can see that the
GC actually has a tristate.
Let's consolidate two variables into one enum.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-3-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
__unix_walk_scc() and unix_walk_scc_fast() call unix_scc_cyclic()
for each SCC to check if it forms a cyclic reference, so that we
can skip GC at the following invocations in case all SCCs do not
have any cycles.
If we count the number of cyclic SCCs in __unix_walk_scc(), we can
simplify unix_walk_scc_fast() because the number of cyclic SCCs
only changes when it garbage-collects a SCC.
So, let's count cyclic SCC in __unix_walk_scc() and decrement it
in unix_walk_scc_fast() when performing garbage collection.
Note that we will use this counter in a later patch to check if a
cycle existed in the previous GC run.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251115020935.2643121-2-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Tariq Toukan says:
====================
net/mlx5: misc changes 2025-11-17
This series contains misc enhancements to the mlx5 driver.
====================
Link: https://patch.msgid.link/1763415729-1238421-1-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Per Documentation/dev-tools/checkpatch.rst, ENOTSUPP is not a standard
error code and should be avoided. EOPNOTSUPP should be used instead.
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Link: https://patch.msgid.link/1763415729-1238421-6-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
In case of a FW issue, FW might be not responding to FW commands,
causing kernel lockout for a long period of time, e.g. rtnl_lock held
while ethtool is trying to collect stats waiting for FW to respond to
multiple commands, when all of them will timeout.
While there's no immediate indication of the FW lockout, we can safely
assume that something is wrong when all command slots are busy and in
a timeout state and no FW completion was received on any of them.
In such case, start immediately failing new commands.
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763415729-1238421-5-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Remove unnecessary logic that sets bw_share to minimal value, when
parent has bw_share configured but nodes don't have min_rate.
This check is redundant because the parent bandwidth acts as the upper
bound regardless, and the firmware always enforces the topmost
bandwidth constraint.
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Cosmin Ratiu <cratiu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763415729-1238421-4-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Extend the TX timestamp handler to recover the SQ when the difference
between the port and CQE TX timestamps is abnormally large.
The current logic aborts timestamp delivery if the delta exceeds
1/128 seconds, which matches the maximum expected packet interval in
ptp4l. A larger delta makes the timestamps unreliable.
This change adds recovery if the delta exceeds 0.5 seconds. Such a
large gap should not occur in normal operation and indicates that
firmware is stuck or metadata tracking is out of sync, leading to stale
or mismatched timestamps. Recovering the SQ ensures forward progress
and avoids silently dropping invalid timestamps.
The timestamp handler now takes mlx5e_ptpsq directly to access both CQ
stats and the recovery state.
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Shahar Shitrit <shshitrit@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763415729-1238421-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Matthew and Jakub reported [1] issues where inventory automation tools
are calling EEPROM query repeatedly on a port that doesn't have an SFP
connected, resulting in millions of error prints.
Move MCIA register status extraction from the query functions to the
callers, allowing use of extack reporting instead of a dmesg print when
using the netlink API.
[1] https://lore.kernel.org/netdev/20251028194011.39877-1-mattc@purestorage.com/
Cc: Matthew W Carlis <mattc@purestorage.com>
Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1763415729-1238421-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Since commit 1b255e1beabf ("tools: ynl: add ipv4-or-v6 display hint"), we
can display either IPv4 or IPv6 addresses for a single field based on the
address family. However, most dual-stack fields still use the ipv4 display
hint. This update changes them to use the new ipv4-or-v6 display hint and
converts IPv4-only fields to use the u32 type.
Field changes:
- v4-or-v6
- IFA_ADDRESS, IFA_LOCAL
- IFLA_GRE_LOCAL, IFLA_GRE_REMOTE
- IFLA_VTI_LOCAL, IFLA_VTI_REMOTE
- IFLA_IPTUN_LOCAL, IFLA_IPTUN_REMOTE
- NDA_DST
- RTA_DST, RTA_SRC, RTA_GATEWAY, RTA_PREFSRC
- FRA_SRC, FRA_DST
- ipv4
- IFA_BROADCAST
- IFLA_GENEVE_REMOTE
- IFLA_IPTUN_6RD_RELAY_PREFIX
Reviewed-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Donald Hunter <donald.hunter@gmail.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://patch.msgid.link/20251117024457.3034-3-liuhangbin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add missing support for parsing MAC addresses when display_hint is 'mac'
in the YNL library. This enables YNL CLI to accept MAC address strings
for attributes like lladdr in rt-neigh operations.
Reviewed-by: Donald Hunter <donald.hunter@gmail.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://patch.msgid.link/20251117024457.3034-2-liuhangbin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Eric Dumazet says:
====================
net: expand napi_skb_cache use
This is a followup of commit e20dfbad8aab ("net: fix napi_consume_skb()
with alien skbs").
Now the per-cpu napi_skb_cache is populated from TX completion path,
we can make use of this cache, especially for cpus not used
from a driver NAPI poll (primary user of napi_cache).
With this series, I consistently reach 130 Mpps on my UDP tx stress test
and reduce SLUB spinlock contention to smaller values.
====================
Link: https://patch.msgid.link/20251116202717.1542829-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This is a followup of commit e20dfbad8aab ("net: fix napi_consume_skb()
with alien skbs").
Now the per-cpu napi_skb_cache is populated from TX completion path,
we can make use of this cache, especially for cpus not used
from a driver NAPI poll (primary user of napi_cache).
We can use the napi_skb_cache only if current context is not from hard irq.
With this patch, I consistently reach 130 Mpps on my UDP tx stress test
and reduce SLUB spinlock contention to smaller values.
Note there is still some SLUB contention for skb->head allocations.
I had to tune /sys/kernel/slab/skbuff_small_head/cpu_partial
and /sys/kernel/slab/skbuff_small_head/min_partial depending
on the platform taxonomy.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Tested-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251116202717.1542829-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This patch refactors __alloc_skb() to prepare the following one,
and does not change functionality.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251116202717.1542829-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
We want to be able in the series last patch to get an skb from
napi_skb_cache from process context, if there is one available.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251116202717.1542829-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
