ocelot_fdma_receive_skb should return false if an unexpected
value is returned by pskb_trim.
Signed-off-by: Yuanjun Gong <ruc_gongyuanjun@163.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
in emac_tso_csum(), return an error code if an unexpected value
is returned by pskb_trim().
Signed-off-by: Yuanjun Gong <ruc_gongyuanjun@163.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2023-07-17 (iavf)
This series contains updates to iavf driver only.
Ding Hui fixes use-after-free issue by calling netif_napi_del() for all
allocated q_vectors. He also resolves out-of-bounds issue by not
updating to new values when timeout is encountered.
Marcin and Ahmed change the way resets are handled so that the callback
operating under the RTNL lock will wait for the reset to finish, the
rtnl_lock sensitive functions in reset flow will schedule the netdev update
for later in order to remove circular dependency with the critical lock.
* '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
iavf: fix reset task race with iavf_remove()
iavf: fix a deadlock caused by rtnl and driver's lock circular dependencies
Revert "iavf: Do not restart Tx queues after reset task failure"
Revert "iavf: Detach device during reset task"
iavf: Wait for reset in callbacks which trigger it
iavf: use internal state to free traffic IRQs
iavf: Fix out-of-bounds when setting channels on remove
iavf: Fix use-after-free in free_netdev
====================
Link: https://lore.kernel.org/r/20230717175205.3217774-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Hardware generated encryption and ICV tags are found to
be wrong when tested with IEEE MACSEC test vectors.
This is because as per the HRM, the hash key (derived by
AES-ECB block encryption of an all 0s block with the SAK)
has to be programmed by the software in
MCSX_RS_MCS_CPM_TX_SLAVE_SA_PLCY_MEM_4X register.
Hence fix this by generating hash key in software and
configuring in hardware.
Fixes: c54ffc7360 ("octeontx2-pf: mcs: Introduce MACSEC hardware offloading")
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Link: https://lore.kernel.org/r/1689574603-28093-1-git-send-email-sbhatta@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In normal operation, each populated queue item has
next_to_watch pointing to the last TX desc of the packet,
while each cleaned item has it set to 0. In particular,
next_to_use that points to the next (necessarily clean)
item to use has next_to_watch set to 0.
When the TX queue is used both by an application using
AF_XDP with ZEROCOPY as well as a second non-XDP application
generating high traffic, the queue pointers can get in
an invalid state where next_to_use points to an item
where next_to_watch is NOT set to 0.
However, the implementation assumes at several places
that this is never the case, so if it does hold,
bad things happen. In particular, within the loop inside
of igc_clean_tx_irq(), next_to_clean can overtake next_to_use.
Finally, this prevents any further transmission via
this queue and it never gets unblocked or signaled.
Secondly, if the queue is in this garbled state,
the inner loop of igc_clean_tx_ring() will never terminate,
completely hogging a CPU core.
The reason is that igc_xdp_xmit_zc() reads next_to_use
before acquiring the lock, and writing it back
(potentially unmodified) later. If it got modified
before locking, the outdated next_to_use is written
pointing to an item that was already used elsewhere
(and thus next_to_watch got written).
Fixes: 9acf59a752 ("igc: Enable TX via AF_XDP zero-copy")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20230717175444.3217831-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2023-07-14 (ice)
This series contains updates to ice driver only.
Petr Oros removes multiple calls made to unregister netdev and
devlink_port.
Michal fixes null pointer dereference that can occur during reload.
====================
Link: https://lore.kernel.org/r/20230714201041.1717834-1-anthony.l.nguyen@intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The reset task is currently scheduled from the watchdog or adminq tasks.
First, all direct calls to schedule the reset task are replaced with the
iavf_schedule_reset(), which is modified to accept the flag showing the
type of reset.
To prevent the reset task from starting once iavf_remove() starts, we need
to check the __IAVF_IN_REMOVE_TASK bit before we schedule it. This is now
easily added to iavf_schedule_reset().
Finally, remove the check for IAVF_FLAG_RESET_NEEDED in the watchdog task.
It is redundant since all callers who set the flag immediately schedules
the reset task.
Fixes: 3ccd54ef44 ("iavf: Fix init state closure on remove")
Fixes: 14756b2ae2 ("iavf: Fix __IAVF_RESETTING state usage")
Signed-off-by: Ahmed Zaki <ahmed.zaki@intel.com>
Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
A driver's lock (crit_lock) is used to serialize all the driver's tasks.
Lockdep, however, shows a circular dependency between rtnl and
crit_lock. This happens when an ndo that already holds the rtnl requests
the driver to reset, since the reset task (in some paths) tries to grab
rtnl to either change real number of queues of update netdev features.
[566.241851] ======================================================
[566.241893] WARNING: possible circular locking dependency detected
[566.241936] 6.2.14-100.fc36.x86_64+debug #1 Tainted: G OE
[566.241984] ------------------------------------------------------
[566.242025] repro.sh/2604 is trying to acquire lock:
[566.242061] ffff9280fc5ceee8 (&adapter->crit_lock){+.+.}-{3:3}, at: iavf_close+0x3c/0x240 [iavf]
[566.242167]
but task is already holding lock:
[566.242209] ffffffff9976d350 (rtnl_mutex){+.+.}-{3:3}, at: iavf_remove+0x6b5/0x730 [iavf]
[566.242300]
which lock already depends on the new lock.
[566.242353]
the existing dependency chain (in reverse order) is:
[566.242401]
-> #1 (rtnl_mutex){+.+.}-{3:3}:
[566.242451] __mutex_lock+0xc1/0xbb0
[566.242489] iavf_init_interrupt_scheme+0x179/0x440 [iavf]
[566.242560] iavf_watchdog_task+0x80b/0x1400 [iavf]
[566.242627] process_one_work+0x2b3/0x560
[566.242663] worker_thread+0x4f/0x3a0
[566.242696] kthread+0xf2/0x120
[566.242730] ret_from_fork+0x29/0x50
[566.242763]
-> #0 (&adapter->crit_lock){+.+.}-{3:3}:
[566.242815] __lock_acquire+0x15ff/0x22b0
[566.242869] lock_acquire+0xd2/0x2c0
[566.242901] __mutex_lock+0xc1/0xbb0
[566.242934] iavf_close+0x3c/0x240 [iavf]
[566.242997] __dev_close_many+0xac/0x120
[566.243036] dev_close_many+0x8b/0x140
[566.243071] unregister_netdevice_many_notify+0x165/0x7c0
[566.243116] unregister_netdevice_queue+0xd3/0x110
[566.243157] iavf_remove+0x6c1/0x730 [iavf]
[566.243217] pci_device_remove+0x33/0xa0
[566.243257] device_release_driver_internal+0x1bc/0x240
[566.243299] pci_stop_bus_device+0x6c/0x90
[566.243338] pci_stop_and_remove_bus_device+0xe/0x20
[566.243380] pci_iov_remove_virtfn+0xd1/0x130
[566.243417] sriov_disable+0x34/0xe0
[566.243448] ice_free_vfs+0x2da/0x330 [ice]
[566.244383] ice_sriov_configure+0x88/0xad0 [ice]
[566.245353] sriov_numvfs_store+0xde/0x1d0
[566.246156] kernfs_fop_write_iter+0x15e/0x210
[566.246921] vfs_write+0x288/0x530
[566.247671] ksys_write+0x74/0xf0
[566.248408] do_syscall_64+0x58/0x80
[566.249145] entry_SYSCALL_64_after_hwframe+0x72/0xdc
[566.249886]
other info that might help us debug this:
[566.252014] Possible unsafe locking scenario:
[566.253432] CPU0 CPU1
[566.254118] ---- ----
[566.254800] lock(rtnl_mutex);
[566.255514] lock(&adapter->crit_lock);
[566.256233] lock(rtnl_mutex);
[566.256897] lock(&adapter->crit_lock);
[566.257388]
*** DEADLOCK ***
The deadlock can be triggered by a script that is continuously resetting
the VF adapter while doing other operations requiring RTNL, e.g:
while :; do
ip link set $VF up
ethtool --set-channels $VF combined 2
ip link set $VF down
ip link set $VF up
ethtool --set-channels $VF combined 4
ip link set $VF down
done
Any operation that triggers a reset can substitute "ethtool --set-channles"
As a fix, add a new task "finish_config" that do all the work which
needs rtnl lock. With the exception of iavf_remove(), all work that
require rtnl should be called from this task.
As for iavf_remove(), at the point where we need to call
unregister_netdevice() (and grab rtnl_lock), we make sure the finish_config
task is not running (cancel_work_sync()) to safely grab rtnl. Subsequent
finish_config work cannot restart after that since the task is guarded
by the __IAVF_IN_REMOVE_TASK bit in iavf_schedule_finish_config().
Fixes: 5ac49f3c27 ("iavf: use mutexes for locking of critical sections")
Signed-off-by: Ahmed Zaki <ahmed.zaki@intel.com>
Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
This reverts commit aa626da947.
Detaching device during reset was not fully fixing the rtnl locking issue,
as there could be a situation where callback was already in progress before
detaching netdev.
Furthermore, detaching netdevice causes TX timeouts if traffic is running.
To reproduce:
ip netns exec ns1 iperf3 -c $PEER_IP -t 600 --logfile /dev/null &
while :; do
for i in 200 7000 400 5000 300 3000 ; do
ip netns exec ns1 ip link set $VF1 mtu $i
sleep 2
done
sleep 10
done
Currently, callbacks such as iavf_change_mtu() wait for the reset.
If the reset fails to acquire the rtnl_lock, they schedule the netdev
update for later while continuing the reset flow. Operations like MTU
changes are performed under the rtnl_lock. Therefore, when the operation
finishes, another callback that uses rtnl_lock can start.
Signed-off-by: Dawid Wesierski <dawidx.wesierski@intel.com>
Signed-off-by: Marcin Szycik <marcin.szycik@linux.intel.com>
Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
There was a fail when trying to add the interface to bonding
right after changing the MTU on the interface. It was caused
by bonding interface unable to open the interface due to
interface being in __RESETTING state because of MTU change.
Add new reset_waitqueue to indicate that reset has finished.
Add waiting for reset to finish in callbacks which trigger hw reset:
iavf_set_priv_flags(), iavf_change_mtu() and iavf_set_ringparam().
We use a 5000ms timeout period because on Hyper-V based systems,
this operation takes around 3000-4000ms. In normal circumstances,
it doesn't take more than 500ms to complete.
Add a function iavf_wait_for_reset() to reuse waiting for reset code and
use it also in iavf_set_channels(), which already waits for reset.
We don't use error handling in iavf_set_channels() as this could
cause the device to be in incorrect state if the reset was scheduled
but hit timeout or the waitng function was interrupted by a signal.
Fixes: 4e5e6b5d9d ("iavf: Fix return of set the new channel count")
Signed-off-by: Marcin Szycik <marcin.szycik@linux.intel.com>
Co-developed-by: Dawid Wesierski <dawidx.wesierski@intel.com>
Signed-off-by: Dawid Wesierski <dawidx.wesierski@intel.com>
Signed-off-by: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
Signed-off-by: Kamil Maziarz <kamil.maziarz@intel.com>
Signed-off-by: Mateusz Palczewski <mateusz.palczewski@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
If the system tries to close the netdev while iavf_reset_task() is
running, __LINK_STATE_START will be cleared and netif_running() will
return false in iavf_reinit_interrupt_scheme(). This will result in
iavf_free_traffic_irqs() not being called and a leak as follows:
[7632.489326] remove_proc_entry: removing non-empty directory 'irq/999', leaking at least 'iavf-enp24s0f0v0-TxRx-0'
[7632.490214] WARNING: CPU: 0 PID: 10 at fs/proc/generic.c:718 remove_proc_entry+0x19b/0x1b0
is shown when pci_disable_msix() is later called. Fix by using the
internal adapter state. The traffic IRQs will always exist if
state == __IAVF_RUNNING.
Fixes: 5b36e8d04b ("i40evf: Enable VF to request an alternate queue allocation")
Signed-off-by: Ahmed Zaki <ahmed.zaki@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Referenced commit missed that for chip versions 42 and 43 ASPM
remained disabled in the respective rtl_hw_start_...() routines.
This resulted in problems as described in the referenced bug
ticket. Therefore re-instantiate the previous logic.
Fixes: 5fc3f6c90c ("r8169: consolidate disabling ASPM before EPHY access")
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217635
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit 6624e780a5 ("ice: split ice_vsi_setup into smaller
functions") ice_vsi_release does things twice. There is unregister
netdev which is unregistered in ice_deinit_eth also.
It also unregisters the devlink_port twice which is also unregistered
in ice_deinit_eth(). This double deregistration is hidden because
devl_port_unregister ignores the return value of xa_erase.
[ 68.642167] Call Trace:
[ 68.650385] ice_devlink_destroy_pf_port+0xe/0x20 [ice]
[ 68.655656] ice_vsi_release+0x445/0x690 [ice]
[ 68.660147] ice_deinit+0x99/0x280 [ice]
[ 68.664117] ice_remove+0x1b6/0x5c0 [ice]
[ 171.103841] Call Trace:
[ 171.109607] ice_devlink_destroy_pf_port+0xf/0x20 [ice]
[ 171.114841] ice_remove+0x158/0x270 [ice]
[ 171.118854] pci_device_remove+0x3b/0xc0
[ 171.122779] device_release_driver_internal+0xc7/0x170
[ 171.127912] driver_detach+0x54/0x8c
[ 171.131491] bus_remove_driver+0x77/0xd1
[ 171.135406] pci_unregister_driver+0x2d/0xb0
[ 171.139670] ice_module_exit+0xc/0x55f [ice]
Fixes: 6624e780a5 ("ice: split ice_vsi_setup into smaller functions")
Signed-off-by: Petr Oros <poros@redhat.com>
Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Identify issues that arise by using the tests/doublebitand.cocci
semantic patch. Need to remove duplicate expression in if statement.
Signed-off-by: Wang Ming <machel@vivo.com>
Reviewed-by: Jiawen Wu <jiawenwu@trustnetic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is expected that most callers should _ignore_ the errors return by
debugfs_create_dir() in bnad_debugfs_init().
Signed-off-by: Wang Ming <machel@vivo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the call to of_get_ethdev_address to mtk_add_mac which is part of
the probe function and can hence itself return -EPROBE_DEFER should
of_get_ethdev_address return -EPROBE_DEFER. This allows us to entirely
get rid of the mtk_init function.
The problem of of_get_ethdev_address returning -EPROBE_DEFER surfaced
in situations in which the NVMEM provider holding the MAC address has
not yet be loaded at the time mtk_eth_soc is initially probed. In this
case probing of mtk_eth_soc should be deferred instead of falling back
to use a random MAC address, so once the NVMEM provider becomes
available probing can be repeated.
Fixes: 656e705243 ("net-next: mediatek: add support for MT7623 ethernet")
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
CPSW ALE has 75 bit ALE entries which are stored within three 32 bit words.
The cpsw_ale_get_field() and cpsw_ale_set_field() functions assume that the
field will be strictly contained within one word. However, this is not
guaranteed to be the case and it is possible for ALE field entries to span
across up to two words at the most.
Fix the methods to handle getting/setting fields spanning up to two words.
Fixes: db82173f23 ("netdev: driver: ethernet: add cpsw address lookup engine support")
Signed-off-by: Tanmay Patil <t-patil@ti.com>
[s-vadapalli@ti.com: rephrased commit message and added Fixes tag]
Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For some device types like TXGBE_ID_XAUI, *checksum computed in
txgbe_calc_eeprom_checksum() is larger than TXGBE_EEPROM_SUM. Remove the
limit on the size of *checksum.
Fixes: 049fe53653 ("net: txgbe: Add operations to interact with firmware")
Fixes: 5e2ea7801f ("net: txgbe: Fix unsigned comparison to zero in txgbe_calc_eeprom_checksum()")
Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
Link: https://lore.kernel.org/r/20230711063414.3311-1-jiawenwu@trustnetic.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The ENA adapters on our instances occasionally reset. Once recently
logged a UBSAN failure to console in the process:
UBSAN: shift-out-of-bounds in build/linux/drivers/net/ethernet/amazon/ena/ena_com.c:540:13
shift exponent 32 is too large for 32-bit type 'unsigned int'
CPU: 28 PID: 70012 Comm: kworker/u72:2 Kdump: loaded not tainted 5.15.117
Hardware name: Amazon EC2 c5d.9xlarge/, BIOS 1.0 10/16/2017
Workqueue: ena ena_fw_reset_device [ena]
Call Trace:
<TASK>
dump_stack_lvl+0x4a/0x63
dump_stack+0x10/0x16
ubsan_epilogue+0x9/0x36
__ubsan_handle_shift_out_of_bounds.cold+0x61/0x10e
? __const_udelay+0x43/0x50
ena_delay_exponential_backoff_us.cold+0x16/0x1e [ena]
wait_for_reset_state+0x54/0xa0 [ena]
ena_com_dev_reset+0xc8/0x110 [ena]
ena_down+0x3fe/0x480 [ena]
ena_destroy_device+0xeb/0xf0 [ena]
ena_fw_reset_device+0x30/0x50 [ena]
process_one_work+0x22b/0x3d0
worker_thread+0x4d/0x3f0
? process_one_work+0x3d0/0x3d0
kthread+0x12a/0x150
? set_kthread_struct+0x50/0x50
ret_from_fork+0x22/0x30
</TASK>
Apparently, the reset delays are getting so large they can trigger a
UBSAN panic.
Looking at the code, the current timeout is capped at 5000us. Using a
base value of 100us, the current code will overflow after (1<<29). Even
at values before 32, this function wraps around, perhaps
unintentionally.
Cap the value of the exponent used for this backoff at (1<<16) which is
larger than currently necessary, but large enough to support bigger
values in the future.
Cc: stable@vger.kernel.org
Fixes: 4bb7f4cf60 ("net: ena: reduce driver load time")
Signed-off-by: Krister Johansen <kjlx@templeofstupid.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Shay Agroskin <shayagr@amazon.com>
Link: https://lore.kernel.org/r/20230711013621.GE1926@templeofstupid.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
igc: Fix corner cases for TSN offload
Florian Kauer says:
The igc driver supports several different offloading capabilities
relevant in the TSN context. Recent patches in this area introduced
regressions for certain corner cases that are fixed in this series.
Each of the patches (except the first one) addresses a different
regression that can be separately reproduced. Still, they have
overlapping code changes so they should not be separately applied.
Especially #4 and #6 address the same observation,
but both need to be applied to avoid TX hang occurrences in
the scenario described in the patches.
====================
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Reviewed-by: Muhammad Husaini Zulkifli <muhammad.husaini.zulkifli@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the case of heavy XDP traffic to be transmitted, the console
will print the error log continuously if there are lack of enough
BDs to accommodate the frames. The log looks like below.
[ 160.013112] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.023116] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.028926] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.038946] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.044758] fec 30be0000.ethernet eth0: NOT enough BD for SG!
Not only will this log be replicated and redundant, it will also
degrade XDP performance. So we use netdev_err_once() instead of
netdev_err() now.
Fixes: 6d6b39f180 ("net: fec: add initial XDP support")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When the XDP feature is enabled and with heavy XDP frames to be
transmitted, there is a considerable probability that available
tx BDs are insufficient. This will lead to some XDP frames to be
discarded and the "NOT enough BD for SG!" error log will appear
in the console (as shown below).
[ 160.013112] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.023116] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.028926] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.038946] fec 30be0000.ethernet eth0: NOT enough BD for SG!
[ 160.044758] fec 30be0000.ethernet eth0: NOT enough BD for SG!
In the case of heavy XDP traffic, sometimes the speed of recycling
tx BDs may be slower than the speed of sending XDP frames. There
may be several specific reasons, such as the interrupt is not
responsed in time, the efficiency of the NAPI callback function is
too low due to all the queues (tx queues and rx queues) share the
same NAPI, and so on.
After trying various methods, I think that increase the size of tx
BD ring is simple and effective. Maybe the best resolution is that
allocate NAPI for each queue to improve the efficiency of the NAPI
callback, but this change is a bit big and I didn't try this method.
Perheps this method will be implemented in a future patch.
This patch also updates the tx_wake_threshold of tx ring which is
related to the size of tx ring in the previous logic. Otherwise,
the tx_wake_threshold will be too high (403 BDs), which is more
likely to impact the slow path in the case of heavy XDP traffic,
because XDP path and slow path share the tx BD rings. According
to Jakub's suggestion, the tx_wake_threshold is at least equal to
tx_stop_threshold + 2 * MAX_SKB_FRAGS, if a queue of hundreds of
entries is overflowing, we should be able to apply a hysteresis
of a few tens of entries.
Fixes: 6d6b39f180 ("net: fec: add initial XDP support")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Once the XDP frames have been successfully transmitted through the
ndo_xdp_xmit() interface, it's the driver responsibility to free
the frames so that the page_pool can recycle the pages and reuse
them. However, this action is not implemented in the fec driver.
This leads to a user-visible problem that the console will print
the following warning log.
[ 157.568851] page_pool_release_retry() stalled pool shutdown 1389 inflight 60 sec
[ 217.983446] page_pool_release_retry() stalled pool shutdown 1389 inflight 120 sec
[ 278.399006] page_pool_release_retry() stalled pool shutdown 1389 inflight 181 sec
[ 338.812885] page_pool_release_retry() stalled pool shutdown 1389 inflight 241 sec
[ 399.226946] page_pool_release_retry() stalled pool shutdown 1389 inflight 302 sec
Therefore, to solve this issue, we free XDP frames via xdp_return_frame()
while cleaning the tx BD ring.
Fixes: 6d6b39f180 ("net: fec: add initial XDP support")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When a XDP program is installed or uninstalled, fec_restart() will
be invoked to reset MAC and buffer descriptor rings. It's reasonable
not to transmit any packet during the process of reset. However, the
NETDEV_XDP_ACT_NDO_XMIT bit of xdp_features is enabled by default,
that is to say, it's possible that the fec_enet_xdp_xmit() will be
invoked even if the process of reset is not finished. In this case,
the redirected XDP frames might be dropped and available transmit BDs
may be incorrectly deemed insufficient. So this patch disable the
NETDEV_XDP_ACT_NDO_XMIT feature by default and dynamically configure
this feature when the bpf program is installed or uninstalled.
Fixes: e4ac7cc6e5 ("net: fec: turn on XDP features")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The insertion of an empty frame was introduced with
commit db0b124f02 ("igc: Enhance Qbv scheduling by using first flag bit")
in order to ensure that the current cycle has at least one packet if
there is some packet to be scheduled for the next cycle.
However, the current implementation does not properly check if
a packet is already scheduled for the current cycle. Currently,
an empty packet is always inserted if and only if
txtime >= end_of_cycle && txtime > last_tx_cycle
but since last_tx_cycle is always either the end of the current
cycle (end_of_cycle) or the end of a previous cycle, the
second part (txtime > last_tx_cycle) is always true unless
txtime == last_tx_cycle.
What actually needs to be checked here is if the last_tx_cycle
was already written within the current cycle, so an empty frame
should only be inserted if and only if
txtime >= end_of_cycle && end_of_cycle > last_tx_cycle.
This patch does not only avoid an unnecessary insertion, but it
can actually be harmful to insert an empty packet if packets
are already scheduled in the current cycle, because it can lead
to a situation where the empty packet is actually processed
as the first packet in the upcoming cycle shifting the packet
with the first_flag even one cycle into the future, finally leading
to a TX hang.
The TX hang can be reproduced on a i225 with:
sudo tc qdisc replace dev enp1s0 parent root handle 100 taprio \
num_tc 1 \
map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \
queues 1@0 \
base-time 0 \
sched-entry S 01 300000 \
flags 0x1 \
txtime-delay 500000 \
clockid CLOCK_TAI
sudo tc qdisc replace dev enp1s0 parent 100:1 etf \
clockid CLOCK_TAI \
delta 500000 \
offload \
skip_sock_check
and traffic generator
sudo trafgen -i traffic.cfg -o enp1s0 --cpp -n0 -q -t1400ns
with traffic.cfg
#define ETH_P_IP 0x0800
{
/* Ethernet Header */
0x30, 0x1f, 0x9a, 0xd0, 0xf0, 0x0e, # MAC Dest - adapt as needed
0x24, 0x5e, 0xbe, 0x57, 0x2e, 0x36, # MAC Src - adapt as needed
const16(ETH_P_IP),
/* IPv4 Header */
0b01000101, 0, # IPv4 version, IHL, TOS
const16(1028), # IPv4 total length (UDP length + 20 bytes (IP header))
const16(2), # IPv4 ident
0b01000000, 0, # IPv4 flags, fragmentation off
64, # IPv4 TTL
17, # Protocol UDP
csumip(14, 33), # IPv4 checksum
/* UDP Header */
10, 0, 48, 1, # IP Src - adapt as needed
10, 0, 48, 10, # IP Dest - adapt as needed
const16(5555), # UDP Src Port
const16(6666), # UDP Dest Port
const16(1008), # UDP length (UDP header 8 bytes + payload length)
csumudp(14, 34), # UDP checksum
/* Payload */
fill('W', 1000),
}
and the observed message with that is for example
igc 0000:01:00.0 enp1s0: Detected Tx Unit Hang
Tx Queue <0>
TDH <32>
TDT <3c>
next_to_use <3c>
next_to_clean <32>
buffer_info[next_to_clean]
time_stamp <ffff26a8>
next_to_watch <00000000632a1828>
jiffies <ffff27f8>
desc.status <1048000>
Fixes: db0b124f02 ("igc: Enhance Qbv scheduling by using first flag bit")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
It is possible (verified on a running system) that frames are processed
by igc_tx_launchtime with a txtime before the start of the cycle
(baset_est).
However, the result of txtime - baset_est is written into a u32,
leading to a wrap around to a positive number. The following
launchtime > 0 check will only branch to executing launchtime = 0
if launchtime is already 0.
Fix it by using a s32 before checking launchtime > 0.
Fixes: db0b124f02 ("igc: Enhance Qbv scheduling by using first flag bit")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The flags IGC_TXQCTL_STRICT_CYCLE and IGC_TXQCTL_STRICT_END
prevent the packet transmission over slot and cycle boundaries.
This is important for taprio offload where the slots and
cycles correspond to the slots and cycles configured for the
network.
However, the Qbv offload feature of the i225 is also used for
enabling TX launchtime / ETF offload. In that case, however,
the cycle has no meaning for the network and is only used
internally to adapt the base time register after a second has
passed.
Enabling strict mode in this case would unnecessarily prevent
the transmission of certain packets (i.e. at the boundary of a
second) and thus interferes with the ETF qdisc that promises
transmission at a certain point in time.
Similar to ETF, this also applies to CBS offload that also should
not be influenced by strict mode unless taprio offload would be
enabled at the same time.
This fully reverts
commit d8f45be01d ("igc: Use strict cycles for Qbv scheduling")
but its commit message only describes what was already implemented
before that commit. The difference to a plain revert of that commit
is that it now copes with the base_time = 0 case that was fixed with
commit e17090eb24 ("igc: allow BaseTime 0 enrollment for Qbv")
In particular, enabling strict mode leads to TX hang situations
under high traffic if taprio is applied WITHOUT taprio offload
but WITH ETF offload, e.g. as in
sudo tc qdisc replace dev enp1s0 parent root handle 100 taprio \
num_tc 1 \
map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \
queues 1@0 \
base-time 0 \
sched-entry S 01 300000 \
flags 0x1 \
txtime-delay 500000 \
clockid CLOCK_TAI
sudo tc qdisc replace dev enp1s0 parent 100:1 etf \
clockid CLOCK_TAI \
delta 500000 \
offload \
skip_sock_check
and traffic generator
sudo trafgen -i traffic.cfg -o enp1s0 --cpp -n0 -q -t1400ns
with traffic.cfg
#define ETH_P_IP 0x0800
{
/* Ethernet Header */
0x30, 0x1f, 0x9a, 0xd0, 0xf0, 0x0e, # MAC Dest - adapt as needed
0x24, 0x5e, 0xbe, 0x57, 0x2e, 0x36, # MAC Src - adapt as needed
const16(ETH_P_IP),
/* IPv4 Header */
0b01000101, 0, # IPv4 version, IHL, TOS
const16(1028), # IPv4 total length (UDP length + 20 bytes (IP header))
const16(2), # IPv4 ident
0b01000000, 0, # IPv4 flags, fragmentation off
64, # IPv4 TTL
17, # Protocol UDP
csumip(14, 33), # IPv4 checksum
/* UDP Header */
10, 0, 48, 1, # IP Src - adapt as needed
10, 0, 48, 10, # IP Dest - adapt as needed
const16(5555), # UDP Src Port
const16(6666), # UDP Dest Port
const16(1008), # UDP length (UDP header 8 bytes + payload length)
csumudp(14, 34), # UDP checksum
/* Payload */
fill('W', 1000),
}
and the observed message with that is for example
igc 0000:01:00.0 enp1s0: Detected Tx Unit Hang
Tx Queue <0>
TDH <d0>
TDT <f0>
next_to_use <f0>
next_to_clean <d0>
buffer_info[next_to_clean]
time_stamp <ffff661f>
next_to_watch <00000000245a4efb>
jiffies <ffff6e48>
desc.status <1048000>
Fixes: d8f45be01d ("igc: Use strict cycles for Qbv scheduling")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Since commit e17090eb24 ("igc: allow BaseTime 0 enrollment for Qbv")
it is possible to enable taprio offload with a basetime of 0.
However, the check if taprio offload is already enabled (and thus -EALREADY
should be returned for igc_save_qbv_schedule) still relied on
adapter->base_time > 0.
This can be reproduced as follows:
# TAPRIO offload (flags == 0x2) and base-time = 0
sudo tc qdisc replace dev enp1s0 parent root handle 100 stab overhead 24 taprio \
num_tc 1 \
map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \
queues 1@0 \
base-time 0 \
sched-entry S 01 300000 \
flags 0x2
# The second call should fail with "Error: Device failed to setup taprio offload."
# But that only happens if base-time was != 0
sudo tc qdisc replace dev enp1s0 parent root handle 100 stab overhead 24 taprio \
num_tc 1 \
map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \
queues 1@0 \
base-time 0 \
sched-entry S 01 300000 \
flags 0x2
Fixes: e17090eb24 ("igc: allow BaseTime 0 enrollment for Qbv")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Only set adapter->taprio_offload_enable after validating the arguments.
Otherwise, it stays set even if the offload was not enabled.
Since the subsequent code does not get executed in case of invalid
arguments, it will not be read at first.
However, by activating and then deactivating another offload
(e.g. ETF/TX launchtime offload), taprio_offload_enable is read
and erroneously keeps the offload feature of the NIC enabled.
This can be reproduced as follows:
# TAPRIO offload (flags == 0x2) and negative base-time leading to expected -ERANGE
sudo tc qdisc replace dev enp1s0 parent root handle 100 stab overhead 24 taprio \
num_tc 1 \
map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \
queues 1@0 \
base-time -1000 \
sched-entry S 01 300000 \
flags 0x2
# IGC_TQAVCTRL is 0x0 as expected (iomem=relaxed for reading register)
sudo pcimem /sys/bus/pci/devices/0000:01:00.0/resource0 0x3570 w*1
# Activate ETF offload
sudo tc qdisc replace dev enp1s0 parent root handle 6666 mqprio \
num_tc 3 \
map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 \
queues 1@0 1@1 2@2 \
hw 0
sudo tc qdisc add dev enp1s0 parent 6666:1 etf \
clockid CLOCK_TAI \
delta 500000 \
offload
# IGC_TQAVCTRL is 0x9 as expected
sudo pcimem /sys/bus/pci/devices/0000:01:00.0/resource0 0x3570 w*1
# Deactivate ETF offload again
sudo tc qdisc delete dev enp1s0 parent 6666:1
# IGC_TQAVCTRL should now be 0x0 again, but is observed as 0x9
sudo pcimem /sys/bus/pci/devices/0000:01:00.0/resource0 0x3570 w*1
Fixes: e17090eb24 ("igc: allow BaseTime 0 enrollment for Qbv")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
In the current implementation the flags adapter->qbv_enable
and IGC_FLAG_TSN_QBV_ENABLED have a similar name, but do not
have the same meaning. The first one is used only to indicate
taprio offload (i.e. when igc_save_qbv_schedule was called),
while the second one corresponds to the Qbv mode of the hardware.
However, the second one is also used to support the TX launchtime
feature, i.e. ETF qdisc offload. This leads to situations where
adapter->qbv_enable is false, but the flag IGC_FLAG_TSN_QBV_ENABLED
is set. This is prone to confusion.
The rename should reduce this confusion. Since it is a pure
rename, it has no impact on functionality.
Fixes: e17090eb24 ("igc: allow BaseTime 0 enrollment for Qbv")
Signed-off-by: Florian Kauer <florian.kauer@linutronix.de>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Current codebase contained the usage of two different names for this
driver (i.e., `gvnic` and `gve`), which is quite unfriendly for users
to use, especially when trying to bind or unbind the driver manually.
The corresponding kernel module is registered with the name of `gve`.
It's more reasonable to align the name of the driver with the module.
Fixes: 893ce44df5 ("gve: Add basic driver framework for Compute Engine Virtual NIC")
Cc: csully@google.com
Signed-off-by: Junfeng Guo <junfeng.guo@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Turning IRQs off is done by accessing Ethernet controller registers.
That can't be done until device's clock is enabled. It results in a SoC
hang otherwise.
This bug remained unnoticed for years as most bootloaders keep all
Ethernet interfaces turned on. It seems to only affect a niche SoC
family BCM47189. It has two Ethernet controllers but CFE bootloader uses
only the first one.
Fixes: 34322615cb ("net: bgmac: Mask interrupts during probe")
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Reviewed-by: Michal Kubiak <michal.kubiak@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the probe error path code that leaves the driver bound
to the device, but with essentially a dead device. This was
useful maybe twice early in the driver's life and no longer
makes sense to keep.
Fixes: 30a1e6d0f8 ("ionic: keep ionic dev on lif init fail")
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove unnecessary early code development check and the WARN_ON
that it uses. The irq alloc and free paths have long been
cleaned up and this check shouldn't have stuck around so long.
Fixes: 77ceb68e29 ("ionic: Add notifyq support")
Signed-off-by: Nitya Sunkad <nitya.sunkad@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Moved PTP pointer validation before its use to avoid smatch warning.
Also used kzalloc/kfree instead of devm_kzalloc/devm_kfree.
Fixes: 2ef4e45d99 ("octeontx2-af: Add PTP PPS Errata workaround on CN10K silicon")
Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: Sai Krishna <saikrishnag@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In legacy silicon, promiscuous mode is only modified
through CGX mbox messages. In CN10KB silicon, it is modified
from CGX mbox and NIX. This breaks legacy application
behaviour. Fix this by removing call from NIX.
Fixes: d6c9784baf ("octeontx2-af: Invoke exact match functions if supported")
Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Michal Kubiak <michal.kubiak@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2023-07-05 (igc)
This series contains updates to igc driver only.
Husaini adds check to increment Qbv change error counter only on taprio
Qbvs. He also removes delay during Tx ring configuration and
resolves Tx hang that could occur when transmitting on a gate to be
closed.
Prasad Koya reports ethtool link mode as TP (twisted pair).
Tee Min corrects value for max SDU.
Aravindhan ensures that registers for PPS are always programmed to occur
in future.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Current duplex mode was unset in the driver, resulting in the default
parameter being set to 0, which corresponds to half duplex. It might
mislead users to have incorrect expectation about the driver's
transmission capabilities.
Set the default duplex configuration to full, as the driver runs in
full duplex mode at this point.
Fixes: 7e074d5a76 ("gve: Enable Link Speed Reporting in the driver.")
Signed-off-by: Junfeng Guo <junfeng.guo@intel.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Message-ID: <20230706044128.2726747-1-junfeng.guo@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2023-07-05 (ice)
This series contains updates to ice driver only.
Sridhar fixes incorrect comparison of max Tx rate limit to occur against
each TC value rather than the aggregate. He also resolves an issue with
the wrong VSI being used when setting max Tx rate when TCs are enabled.
* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
ice: Fix tx queue rate limit when TCs are configured
ice: Fix max_rate check while configuring TX rate limits
====================
Link: https://lore.kernel.org/r/20230705201346.49370-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Saeed Mahameed says:
====================
mlx5 fixes 2023-07-05
This series provides bug fixes to mlx5 driver.
* tag 'mlx5-fixes-2023-07-05' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
net/mlx5e: RX, Fix page_pool page fragment tracking for XDP
net/mlx5: Query hca_cap_2 only when supported
net/mlx5e: TC, CT: Offload ct clear only once
net/mlx5e: Check for NOT_READY flag state after locking
net/mlx5: Register a unique thermal zone per device
net/mlx5e: RX, Fix flush and close release flow of regular rq for legacy rq
net/mlx5e: fix memory leak in mlx5e_ptp_open
net/mlx5e: fix memory leak in mlx5e_fs_tt_redirect_any_create
net/mlx5e: fix double free in mlx5e_destroy_flow_table
====================
Link: https://lore.kernel.org/r/20230705175757.284614-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
