commit 706136c572 upstream.
Add a final dma_wmb() barrier before triggering the transmit request
(TCCR_TSRQ) to ensure all descriptor and buffer writes are visible to
the DMA engine.
According to the hardware manual, a read-back operation is required
before writing to the doorbell register to guarantee completion of
previous writes. Instead of performing a dummy read, a dma_wmb() is
used to both enforce the same ordering semantics on the CPU side and
also to ensure completion of writes.
Fixes: c156633f13 ("Renesas Ethernet AVB driver proper")
Cc: stable@vger.kernel.org
Co-developed-by: Fabrizio Castro <fabrizio.castro.jz@renesas.com>
Signed-off-by: Fabrizio Castro <fabrizio.castro.jz@renesas.com>
Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Link: https://patch.msgid.link/20251017151830.171062-5-prabhakar.mahadev-lad.rj@bp.renesas.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5370c31e84 upstream.
Ensure the TX descriptor type fields are published in a safe order so the
DMA engine never begins processing a descriptor chain before all descriptor
fields are fully initialised.
For multi-descriptor transmits the driver writes DT_FEND into the last
descriptor and DT_FSTART into the first. The DMA engine begins processing
when it observes DT_FSTART. Move the dma_wmb() barrier so it executes
immediately after DT_FEND and immediately before writing DT_FSTART
(and before DT_FSINGLE in the single-descriptor case). This guarantees
that all prior CPU writes to the descriptor memory are visible to the
device before DT_FSTART is seen.
This avoids a situation where compiler/CPU reordering could publish
DT_FSTART ahead of DT_FEND or other descriptor fields, allowing the DMA to
start on a partially initialised chain and causing corrupted transmissions
or TX timeouts. Such a failure was observed on RZ/G2L with an RT kernel as
transmit queue timeouts and device resets.
Fixes: 2f45d1902a ("ravb: minimize TX data copying")
Cc: stable@vger.kernel.org
Co-developed-by: Fabrizio Castro <fabrizio.castro.jz@renesas.com>
Signed-off-by: Fabrizio Castro <fabrizio.castro.jz@renesas.com>
Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Link: https://patch.msgid.link/20251017151830.171062-4-prabhakar.mahadev-lad.rj@bp.renesas.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 7f864458e9 upstream.
On all platforms set_clock_selection() writes to a GRF register. This
requires certain clocks running and thus should happen before the
clocks are disabled.
This has been noticed on RK3576 Sige5, which hangs during system suspend
when trying to suspend the second network interface. Note, that
suspending the first interface works, because the second device ensures
that the necessary clocks for the GRF are enabled.
Cc: stable@vger.kernel.org
Fixes: 2f2b60a0ec ("net: ethernet: stmmac: dwmac-rk: Add gmac support for rk3588")
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251014-rockchip-network-clock-fix-v1-1-c257b4afdf75@collabora.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit d63f0391d6 ]
hibmcge uses fixed_phy_register() et al, but doesn't cater for the case
that hibmcge is built-in and fixed_phy is a module. To solve this
select FIXED_PHY.
Fixes: 1d7cd7a9c6 ("net: hibmcge: support scenario without PHY")
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Jijie Shao <shaojijie@huawei.com>
Link: https://patch.msgid.link/c4fc061f-b6d5-418b-a0dc-6b238cdbedce@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 49d34f3dd8 ]
Resolve race conditions in timestamp events list handling between TX
and RX paths causing missed timestamps.
The current implementation uses a single events list for both TX and RX
timestamps. The am65_cpts_find_ts() function acquires the lock,
splices all events (TX as well as RX events) to a temporary list,
and releases the lock. This function performs matching of timestamps
for TX packets only. Before it acquires the lock again to put the
non-TX events back to the main events list, a concurrent RX
processing thread could acquire the lock (as observed in practice),
find an empty events list, and fail to attach timestamp to it,
even though a relevant event exists in the spliced list which is yet to
be restored to the main list.
Fix this by creating separate events lists to handle TX and RX
timestamps independently.
Fixes: c459f606f6 ("net: ethernet: ti: am65-cpts: Enable RX HW timestamp for PTP packets using CPTS FIFO")
Signed-off-by: Aksh Garg <a-garg7@ti.com>
Reviewed-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Link: https://patch.msgid.link/20251016115755.1123646-1-a-garg7@ti.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 87bcef158a ]
XDP programs can change the layout of an xdp_buff through
bpf_xdp_adjust_tail() and bpf_xdp_adjust_head(). Therefore, the driver
cannot assume the size of the linear data area nor fragments. Fix the
bug in mlx5 by generating skb according to xdp_buff after XDP programs
run.
Currently, when handling multi-buf XDP, the mlx5 driver assumes the
layout of an xdp_buff to be unchanged. That is, the linear data area
continues to be empty and fragments remain the same. This may cause
the driver to generate erroneous skb or triggering a kernel
warning. When an XDP program added linear data through
bpf_xdp_adjust_head(), the linear data will be ignored as
mlx5e_build_linear_skb() builds an skb without linear data and then
pull data from fragments to fill the linear data area. When an XDP
program has shrunk the non-linear data through bpf_xdp_adjust_tail(),
the delta passed to __pskb_pull_tail() may exceed the actual nonlinear
data size and trigger the BUG_ON in it.
To fix the issue, first record the original number of fragments. If the
number of fragments changes after the XDP program runs, rewind the end
fragment pointer by the difference and recalculate the truesize. Then,
build the skb with the linear data area matching the xdp_buff. Finally,
only pull data in if there is non-linear data and fill the linear part
up to 256 bytes.
Fixes: f52ac7028b ("net/mlx5e: RX, Add XDP multi-buffer support in Striding RQ")
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1760644540-899148-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit afd5ba577c ]
XDP programs can release xdp_buff fragments when calling
bpf_xdp_adjust_tail(). The driver currently assumes the number of
fragments to be unchanged and may generate skb with wrong truesize or
containing invalid frags. Fix the bug by generating skb according to
xdp_buff after the XDP program runs.
Fixes: ea5d49bdae ("net/mlx5e: Add XDP multi buffer support to the non-linear legacy RQ")
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1760644540-899148-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 902e81e679 ]
The blamed commit increased the needed headroom to account for
alignment. This means that the size required to always align a Tx buffer
was added inside the dpaa2_eth_needed_headroom() function. By doing
that, a manual adjustment of the pointer passed to PTR_ALIGN() was no
longer correct since the 'buffer_start' variable was already pointing
to the start of the skb's memory.
The behavior of the dpaa2-eth driver without this patch was to drop
frames on Tx even when the headroom was matching the 128 bytes
necessary. Fix this by removing the manual adjust of 'buffer_start' from
the PTR_MODE call.
Closes: https://lore.kernel.org/netdev/70f0dcd9-1906-4d13-82df-7bbbbe7194c6@app.fastmail.com/T/#u
Fixes: f422abe3f2 ("dpaa2-eth: increase the needed headroom to account for alignment")
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Tested-by: Mathew McBride <matt@traverse.com.au>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251016135807.360978-1-ioana.ciornei@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit e59bc32df2 ]
The ENETC RX ring uses the page halves flipping mechanism, each page is
split into two halves for the RX ring to use. And ENETC_RXB_TRUESIZE is
defined to 2048 to indicate the size of half a page. However, the page
size is configurable, for ARM64 platform, PAGE_SIZE is default to 4K,
but it could be configured to 16K or 64K.
When PAGE_SIZE is set to 16K or 64K, ENETC_RXB_TRUESIZE is not correct,
and the RX ring will always use the first half of the page. This is not
consistent with the description in the relevant kernel doc and commit
messages.
This issue is invisible in most cases, but if users want to increase
PAGE_SIZE to receive a Jumbo frame with a single buffer for some use
cases, it will not work as expected, because the buffer size of each
RX BD is fixed to 2048 bytes.
Based on the above two points, we expect to correct ENETC_RXB_TRUESIZE
to (PAGE_SIZE >> 1), as described in the comment.
Fixes: d4fd0404c1 ("enetc: Introduce basic PF and VF ENETC ethernet drivers")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Link: https://patch.msgid.link/20251016080131.3127122-1-wei.fang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 50bd33f6b3 ]
After applying the workaround for err050089, the LS1028A platform
experiences RCU stalls on RT kernel. This issue is caused by the
recursive acquisition of the read lock enetc_mdio_lock. Here list some
of the call stacks identified under the enetc_poll path that may lead to
a deadlock:
enetc_poll
-> enetc_lock_mdio
-> enetc_clean_rx_ring OR napi_complete_done
-> napi_gro_receive
-> enetc_start_xmit
-> enetc_lock_mdio
-> enetc_map_tx_buffs
-> enetc_unlock_mdio
-> enetc_unlock_mdio
After enetc_poll acquires the read lock, a higher-priority writer attempts
to acquire the lock, causing preemption. The writer detects that a
read lock is already held and is scheduled out. However, readers under
enetc_poll cannot acquire the read lock again because a writer is already
waiting, leading to a thread hang.
Currently, the deadlock is avoided by adjusting enetc_lock_mdio to prevent
recursive lock acquisition.
Fixes: 6d36ecdbc4 ("net: enetc: take the MDIO lock only once per NAPI poll cycle")
Signed-off-by: Jianpeng Chang <jianpeng.chang.cn@windriver.com>
Acked-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20251015021427.180757-1-jianpeng.chang.cn@windriver.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit aaf043a568 ]
When building with Clang 20 or newer, there are some objtool warnings
from unexpected fallthroughs to other functions:
vmlinux.o: warning: objtool: mlx5e_mpwrq_mtts_per_wqe() falls through to next function mlx5e_mpwrq_max_num_entries()
vmlinux.o: warning: objtool: mlx5e_mpwrq_max_log_rq_size() falls through to next function mlx5e_get_linear_rq_headroom()
LLVM 20 contains an (admittedly problematic [1]) optimization [2] to
convert divide by zero into the equivalent of __builtin_unreachable(),
which invokes undefined behavior and destroys code generation when it is
encountered in a control flow graph.
mlx5e_mpwrq_umr_entry_size() returns 0 in the default case of an
unrecognized mlx5e_mpwrq_umr_mode value. mlx5e_mpwrq_mtts_per_wqe(),
which is inlined into mlx5e_mpwrq_max_log_rq_size(), uses the result of
mlx5e_mpwrq_umr_entry_size() in a divide operation without checking for
zero, so LLVM is able to infer there will be a divide by zero in this
case and invokes undefined behavior. While there is some proposed work
to isolate this undefined behavior and avoid the destructive code
generation that results in these objtool warnings, code should still be
defensive against divide by zero.
As the WARN_ONCE() implies that an invalid value should be handled
gracefully, return 1 instead of 0 in the default case so that the
results of this division operation is always valid.
Fixes: 168723c1f8 ("net/mlx5e: xsk: Use umr_mode to calculate striding RQ parameters")
Link: https://lore.kernel.org/CAGG=3QUk8-Ak7YKnRziO4=0z=1C_7+4jF+6ZeDQ9yF+kuTOHOQ@mail.gmail.com/ [1]
Link: 37932643ab [2]
Closes: https://github.com/ClangBuiltLinux/linux/issues/2131
Closes: https://github.com/ClangBuiltLinux/linux/issues/2132
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20251014-mlx5e-avoid-zero-div-from-mlx5e_mpwrq_umr_entry_size-v1-1-dc186b8819ef@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0c3f2e6281 ]
Some execution paths that jump to the fiber_setup_done label
could leave the remote_adv and local_adv variables uninitialized
and then use it.
Initialize this variables at the point of definition to avoid this.
Fixes: 85730a631f ("tg3: Add SGMII phy support for 5719/5718 serdes")
Co-developed-by: Alexandr Sapozhnikov <alsp705@gmail.com>
Signed-off-by: Alexandr Sapozhnikov <alsp705@gmail.com>
Signed-off-by: Alexey Simakov <bigalex934@gmail.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://patch.msgid.link/20251014164736.5890-1-bigalex934@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit bd5afca115 ]
Completion napi can free out-of-order tx descriptors if hw QoS is
enabled and packets with different priority are queued to same DMA ring.
Take into account possible out-of-order reports checking if the tx queue
is full using circular buffer head/tail pointer instead of the number of
queued packets.
Fixes: 23020f0493 ("net: airoha: Introduce ethernet support for EN7581 SoC")
Suggested-by: Simon Horman <horms@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251012-airoha-tx-busy-queue-v2-1-a600b08bab2d@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2616222e42 ]
During interface toggle operations (ifdown/ifup), the driver currently
resets the local helper variable 'phy_link' to -1. This causes the link
state machine to incorrectly interpret the state as a link change event,
resulting in spurious "Link is down" messages being logged when the
interface is brought back up.
Preserve the phy_link state across interface toggles to avoid treating
the -1 sentinel value as a legitimate link state transition.
Fixes: 88131a812b ("amd-xgbe: Perform phy connect/disconnect at dev open/stop")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Reviewed-by: Dawid Osuchowski <dawid.osuchowski@linux.intel.com>
Link: https://patch.msgid.link/20251010065142.1189310-1-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a3f8c0a273 ]
When the driver requests Tx timestamp value, one of the first steps is
to clone SKB using skb_get. It increases the reference counter for that
SKB to prevent unexpected freeing by another component.
However, there may be a case where the index is requested, SKB is
assigned and never consumed by PTP flows - for example due to reset during
running PTP apps.
Add a check in release timestamping function to verify if the SKB
assigned to Tx timestamp latch was freed, and release remaining SKBs.
Fixes: 4901e83a94 ("idpf: add Tx timestamp capabilities negotiation")
Signed-off-by: Milena Olech <milena.olech@intel.com>
Signed-off-by: Anton Nadezhdin <anton.nadezhdin@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Tested-by: Samuel Salin <Samuel.salin@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20251009-jk-iwl-net-2025-10-01-v3-1-ef32a425b92a@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 70f92ab970 ]
After resume from S4 (hibernate), RTL8168H/RTL8111H truncates incoming
packets. Packet captures show messages like "IP truncated-ip - 146 bytes
missing!".
The issue is caused by RxConfig not being properly re-initialized after
resume. Re-initializing the RxConfig register before the chip
re-initialization sequence avoids the truncation and restores correct
packet reception.
This follows the same pattern as commit ef9da46dde ("r8169: fix data
corruption issue on RTL8402").
Fixes: 6e1d0b8988 ("r8169:add support for RTL8168H and RTL8107E")
Signed-off-by: Linmao Li <lilinmao@kylinos.cn>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://patch.msgid.link/20251009122549.3955845-1-lilinmao@kylinos.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 65946eac6d ]
There is no error handling for `dma_map_single()` failures.
Add error handling by checking `dma_mapping_error()` and freeing
the `skb` using `dev_kfree_skb()` (process context) when it fails.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Yeounsu Moon <yyyynoom@gmail.com>
Tested-on: D-Link DGE-550T Rev-A3
Suggested-by: Simon Horman <horms@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3abc0e55ea ]
Limit tx/rx buffer address to 32-bit address space for board with more
than 4GB DRAM.
Fixes: 804775dfc2 ("net: ethernet: mtk_eth_soc: add support for Wireless Ethernet Dispatch (WED)")
Fixes: 6757d345dd ("net: ethernet: mtk_wed: introduce hw_rro support for MT7988")
Tested-by: Daniel Pawlik <pawlik.dan@gmail.com>
Tested-by: Matteo Croce <teknoraver@meta.com>
Signed-off-by: Rex Lu <rex.lu@mediatek.com>
Co-developed-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit bfdd74166a upstream.
The device returns a valid bit in the LSB of the low timestamp byte in
the completion descriptor that the driver should check before
setting the SKB's hardware timestamp. If the timestamp is not valid, do not
hardware timestamp the SKB.
Cc: stable@vger.kernel.org
Fixes: b2c7aeb490 ("gve: Implement ndo_hwtstamp_get/set for RX timestamping")
Reviewed-by: Joshua Washington <joshwash@google.com>
Signed-off-by: Tim Hostetler <thostet@google.com>
Signed-off-by: Harshitha Ramamurthy <hramamurthy@google.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20251014004740.2775957-1-hramamurthy@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit a7075f501b upstream.
There was backward compatibility in the terms of mailbox API. Various
drivers from various OSes supporting 10G adapters from Intel portfolio
could easily negotiate mailbox API.
This convention has been broken since introducing API 1.4.
Commit 0062e7cc95 ("ixgbevf: add VF IPsec offload code") added support
for IPSec which is specific only for the kernel ixgbe driver. None of the
rest of the Intel 10G PF/VF drivers supports it. And actually lack of
support was not included in the IPSec implementation - there were no such
code paths. No possibility to negotiate support for the feature was
introduced along with introduction of the feature itself.
Commit 339f289641 ("ixgbevf: Add support for new mailbox communication
between PF and VF") increasing API version to 1.5 did the same - it
introduced code supported specifically by the PF ESX driver. It altered API
version for the VF driver in the same time not touching the version
defined for the PF ixgbe driver. It led to additional discrepancies,
as the code provided within API 1.6 cannot be supported for Linux ixgbe
driver as it causes crashes.
The issue was noticed some time ago and mitigated by Jake within the commit
d0725312ad ("ixgbevf: stop attempting IPSEC offload on Mailbox API 1.5").
As a result we have regression for IPsec support and after increasing API
to version 1.6 ixgbevf driver stopped to support ESX MBX.
To fix this mess add new mailbox op asking PF driver about supported
features. Basing on a response determine whether to set support for IPSec
and ESX-specific enhanced mailbox.
New mailbox op, for compatibility purposes, must be added within new API
revision, as API version of OOT PF & VF drivers is already increased to
1.6 and doesn't incorporate features negotiate op.
Features negotiation mechanism gives possibility to be extended with new
features when needed in the future.
Reported-by: Jacob Keller <jacob.e.keller@intel.com>
Closes: https://lore.kernel.org/intel-wired-lan/20241101-jk-ixgbevf-mailbox-v1-5-fixes-v1-0-f556dc9a66ed@intel.com/
Fixes: 0062e7cc95 ("ixgbevf: add VF IPsec offload code")
Fixes: 339f289641 ("ixgbevf: Add support for new mailbox communication between PF and VF")
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jedrzej Jagielski <jedrzej.jagielski@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20251009-jk-iwl-net-2025-10-01-v3-4-ef32a425b92a@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 53f0eb62b4 upstream.
E610 adapters no longer use the VFLINKS register to read PF's link
speed and linkup state. As a result VF driver cannot get actual link
state and it incorrectly reports 10G which is the default option.
It leads to a situation where even 1G adapters print 10G as actual
link speed. The same happens when PF driver set speed different than 10G.
Add new mailbox operation to let the VF driver request a PF driver
to provide actual link data. Update the mailbox api to v1.6.
Incorporate both ways of getting link status within the legacy
ixgbe_check_mac_link_vf() function.
Fixes: 4c44b450c6 ("ixgbevf: Add support for Intel(R) E610 device")
Co-developed-by: Andrzej Wilczynski <andrzejx.wilczynski@intel.com>
Signed-off-by: Andrzej Wilczynski <andrzejx.wilczynski@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jedrzej Jagielski <jedrzej.jagielski@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20251009-jk-iwl-net-2025-10-01-v3-2-ef32a425b92a@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 22239eb258 ]
When configuring IPsec packet offload in tunnel mode, the driver tries
to create tunnel reformat objects unconditionally. This is incorrect,
because tunnel mode is only permitted under specific encapsulation
settings, and that decision is already made when the flow table is
created.
The offending commit attempted to block this case in the state add
path, but the check there happens too late and does not prevent the
reformat from being configured.
Fix by taking short reservations for both the eswitch mode and the
encap at the start of state setup. This preserves the block ordering
(mode --> encap) used later: the mode is blocked during RX/TX get, and
the encap is blocked during flow-table creation. This lets us fail
early if either reservation cannot be obtained, it means a mode
transition is underway or a conflicting configuration already owns
encap. If both succeed, the flow-table path later takes the ownership
and the reservations are released on exit.
Fixes: 146c196b60 ("net/mlx5e: Create IPsec table with tunnel support only when encap is disabled")
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1759652999-858513-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 7593439c13 ]
When creating IPsec flow tables with tunnel mode enabled, the driver
uses mlx5_eswitch_block_encap() to prevent tunnel encapsulation
conflicts across different domains (NIC_RX/NIC_TX and FDB), since the
firmware doesn’t allow both at the same time.
Currently, the driver attempts to reserve tunnel mode unconditionally
for both NIC and FDB IPsec tables. This can lead to conflicting tunnel
mode setups, for example, if a flow table was created in the FDB
domain with tunnel offload enabled, and we later try to create another
one in the NIC, or vice versa.
To resolve this, adjust the blocking logic so that tunnel mode is only
reserved by NIC flows. This ensures that tunnel offload is exclusively
used in either the NIC or the FDB, and avoids unintended offload
conflicts.
Fixes: 1762f132d5 ("net/mlx5e: Support IPsec packet offload for RX in switchdev mode")
Fixes: c6c2bf5db4 ("net/mlx5e: Support IPsec packet offload for TX in switchdev mode")
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1759652999-858513-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c9d1b0b542 ]
The sparx5 driver programs UC/MC/BC flooding in sparx5_update_fwd() by
unconditionally applying bridge_fwd_mask to all flood PGIDs. Any bridge
topology change that triggers sparx5_update_fwd() (for example enslaving
another port) therefore reinstalls flooding in hardware for already
bridged ports, regardless of their per-port flood flags.
This results in clobbering of the flood masks, and desynchronization
between software and hardware: the bridge still reports “flood off” for
the port, but hardware has flooding enabled due to unconditional PGID
reprogramming.
Steps to reproduce:
$ ip link add br0 type bridge
$ ip link set br0 up
$ ip link set eth0 master br0
$ ip link set eth0 up
$ bridge link set dev eth0 flood off
$ ip link set eth1 master br0
$ ip link set eth1 up
At this point, flooding is silently re-enabled for eth0. Software still
shows “flood off” for eth0, but hardware has flooding enabled.
To fix this, flooding is now set explicitly during bridge join/leave,
through sparx5_port_attr_bridge_flags():
On bridge join, UC/MC/BC flooding is enabled by default.
On bridge leave, UC/MC/BC flooding is disabled.
sparx5_update_fwd() no longer touches the flood PGIDs, clobbering
the flood masks, and desynchronizing software and hardware.
Initialization of the flooding PGIDs have been moved to
sparx5_start(). This is required as flooding PGIDs defaults to
0x3fffffff in hardware and the initialization was previously handled
in sparx5_update_fwd(), which was removed.
With this change, user-configured flooding flags persist across bridge
updates and are no longer overridden by sparx5_update_fwd().
Fixes: d6fce51419 ("net: sparx5: add switching support")
Signed-off-by: Daniel Machon <daniel.machon@microchip.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251003-fix-flood-fwd-v1-1-48eb478b2904@microchip.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2db687f346 ]
When ice_adapter_new() fails, the reserved XArray entry created by
xa_insert() is not released. This causes subsequent insertions at
the same index to return -EBUSY, potentially leading to
NULL pointer dereferences.
Reorder the operations as suggested by Przemek Kitszel:
1. Check if adapter already exists (xa_load)
2. Reserve the XArray slot (xa_reserve)
3. Allocate the adapter (ice_adapter_new)
4. Store the adapter (xa_store)
Fixes: 0f0023c649 ("ice: do not init struct ice_adapter more times than needed")
Suggested-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Suggested-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Haotian Zhang <vulab@iscas.ac.cn>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20251001115336.1707-1-vulab@iscas.ac.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit bc9ea78707 ]
The origin code calls cancel_delayed_work() in ocelot_stats_deinit()
to cancel the cyclic delayed work item ocelot->stats_work. However,
cancel_delayed_work() may fail to cancel the work item if it is already
executing. While destroy_workqueue() does wait for all pending work items
in the work queue to complete before destroying the work queue, it cannot
prevent the delayed work item from being rescheduled within the
ocelot_check_stats_work() function. This limitation exists because the
delayed work item is only enqueued into the work queue after its timer
expires. Before the timer expiration, destroy_workqueue() has no visibility
of this pending work item. Once the work queue appears empty,
destroy_workqueue() proceeds with destruction. When the timer eventually
expires, the delayed work item gets queued again, leading to the following
warning:
workqueue: cannot queue ocelot_check_stats_work on wq ocelot-switch-stats
WARNING: CPU: 2 PID: 0 at kernel/workqueue.c:2255 __queue_work+0x875/0xaf0
...
RIP: 0010:__queue_work+0x875/0xaf0
...
RSP: 0018:ffff88806d108b10 EFLAGS: 00010086
RAX: 0000000000000000 RBX: 0000000000000101 RCX: 0000000000000027
RDX: 0000000000000027 RSI: 0000000000000004 RDI: ffff88806d123e88
RBP: ffffffff813c3170 R08: 0000000000000000 R09: ffffed100da247d2
R10: ffffed100da247d1 R11: ffff88806d123e8b R12: ffff88800c00f000
R13: ffff88800d7285c0 R14: ffff88806d0a5580 R15: ffff88800d7285a0
FS: 0000000000000000(0000) GS:ffff8880e5725000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fe18e45ea10 CR3: 0000000005e6c000 CR4: 00000000000006f0
Call Trace:
<IRQ>
? kasan_report+0xc6/0xf0
? __pfx_delayed_work_timer_fn+0x10/0x10
? __pfx_delayed_work_timer_fn+0x10/0x10
call_timer_fn+0x25/0x1c0
__run_timer_base.part.0+0x3be/0x8c0
? __pfx_delayed_work_timer_fn+0x10/0x10
? rcu_sched_clock_irq+0xb06/0x27d0
? __pfx___run_timer_base.part.0+0x10/0x10
? try_to_wake_up+0xb15/0x1960
? _raw_spin_lock_irq+0x80/0xe0
? __pfx__raw_spin_lock_irq+0x10/0x10
tmigr_handle_remote_up+0x603/0x7e0
? __pfx_tmigr_handle_remote_up+0x10/0x10
? sched_balance_trigger+0x1c0/0x9f0
? sched_tick+0x221/0x5a0
? _raw_spin_lock_irq+0x80/0xe0
? __pfx__raw_spin_lock_irq+0x10/0x10
? tick_nohz_handler+0x339/0x440
? __pfx_tmigr_handle_remote_up+0x10/0x10
__walk_groups.isra.0+0x42/0x150
tmigr_handle_remote+0x1f4/0x2e0
? __pfx_tmigr_handle_remote+0x10/0x10
? ktime_get+0x60/0x140
? lapic_next_event+0x11/0x20
? clockevents_program_event+0x1d4/0x2a0
? hrtimer_interrupt+0x322/0x780
handle_softirqs+0x16a/0x550
irq_exit_rcu+0xaf/0xe0
sysvec_apic_timer_interrupt+0x70/0x80
</IRQ>
...
The following diagram reveals the cause of the above warning:
CPU 0 (remove) | CPU 1 (delayed work callback)
mscc_ocelot_remove() |
ocelot_deinit() | ocelot_check_stats_work()
ocelot_stats_deinit() |
cancel_delayed_work()| ...
| queue_delayed_work()
destroy_workqueue() | (wait a time)
| __queue_work() //UAF
The above scenario actually constitutes a UAF vulnerability.
The ocelot_stats_deinit() is only invoked when initialization
failure or resource destruction, so we must ensure that any
delayed work items cannot be rescheduled.
Replace cancel_delayed_work() with disable_delayed_work_sync()
to guarantee proper cancellation of the delayed work item and
ensure completion of any currently executing work before the
workqueue is deallocated.
A deadlock concern was considered: ocelot_stats_deinit() is called
in a process context and is not holding any locks that the delayed
work item might also need. Therefore, the use of the _sync() variant
is safe here.
This bug was identified through static analysis. To reproduce the
issue and validate the fix, I simulated ocelot-switch device by
writing a kernel module and prepared the necessary resources for
the virtual ocelot-switch device's probe process. Then, removing
the virtual device will trigger the mscc_ocelot_remove() function,
which in turn destroys the workqueue.
Fixes: a556c76adc ("net: mscc: Add initial Ocelot switch support")
Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Link: https://patch.msgid.link/20251001011149.55073-1-duoming@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 4f0d91ba72 ]
Print "entry->mac" before freeing "entry". The "entry" pointer is
freed with kfree_rcu() so it's unlikely that we would trigger this
in real life, but it's safer to re-order it.
Fixes: cc5387f734 ("net/mlx4_en: Add unicast MAC filtering")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/aNvMHX4g8RksFFvV@stanley.mountain
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 92e9f4faff ]
The bitmap allocated with bitmap_zalloc() in otx2_probe() was not
released in otx2_remove(). Unbinding and rebinding the driver therefore
triggers a kmemleak warning:
unreferenced object (size 8):
backtrace:
bitmap_zalloc
otx2_probe
Call bitmap_free() in the remove path to fix the leak.
Fixes: efabce2901 ("octeontx2-pf: AF_XDP zero copy receive support")
Signed-off-by: Bo Sun <bo@mboxify.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit cd9ea7da41 ]
The bitmap allocated with bitmap_zalloc() in otx2vf_probe() was not
released in otx2vf_remove(). Unbinding and rebinding the driver therefore
triggers a kmemleak warning:
unreferenced object (size 8):
backtrace:
bitmap_zalloc
otx2vf_probe
Call bitmap_free() in the remove path to fix the leak.
Fixes: efabce2901 ("octeontx2-pf: AF_XDP zero copy receive support")
Signed-off-by: Bo Sun <bo@mboxify.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 6f5dacf88a ]
This reverts commit ceddedc969.
Commit in question breaks the mapping of PGs to pools for some SKUs.
Specifically multi-host NICs seem to be shipped with a custom buffer
configuration which maps the lossy PG to pool 4. But the bad commit
overrides this with pool 0 which does not have sufficient buffer space
reserved. Resulting in ~40% packet loss. The commit also breaks BMC /
OOB connection completely (100% packet loss).
Revert, similarly to commit 3fbfe251cc ("Revert "net/mlx5e: Update and
set Xon/Xoff upon port speed set""). The breakage is exactly the same,
the only difference is that quoted commit would break the NIC immediately
on boot, and the currently reverted commit only when MTU is changed.
Note: "good" kernels do not restore the configuration, so downgrade isn't
enough to recover machines. A NIC power cycle seems to be necessary to
return to a healthy state (or overriding the relevant registers using
a custom patch).
Fixes: ceddedc969 ("net/mlx5e: Update and set Xon/Xoff upon MTU set")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20250929181529.1848157-1-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2aff4420ef ]
Software can only initialize the PIR and CIR of the command BD ring after
a FLR, and these two registers can only be set to 0. But the reset values
of these two registers are 0, so software does not need to update them.
If there is no a FLR and PIR and CIR are not 0, resetting them to 0 or
other values by software will cause the command BD ring to work
abnormally. This is because of an internal context in the ring prefetch
logic that will retain the state from the first incarnation of the ring
and continue prefetching from the stale location when the ring is
reinitialized. The internal context can only be reset by the FLR.
In addition, there is a logic error in the implementation, next_to_clean
indicates the software CIR and next_to_use indicates the software PIR.
But the current driver uses next_to_clean to set PIR and use next_to_use
to set CIR. This does not cause a problem in actual use, because the
current command BD ring is only initialized after FLR, and the initial
values of next_to_use and next_to_clean are both 0.
Therefore, this patch removes the initialization of PIR and CIR. Instead,
next_to_use and next_to_clean are initialized by reading the values of
PIR and CIR.
Fixes: 4701073c3d ("net: enetc: add initial netc-lib driver to support NTMP")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20250926013954.2003456-1-wei.fang@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 5cfbe7ebfa ]
Add sync reset timeout to stop poll_sync_reset in case there was no
reset done or abort event within timeout. Otherwise poll sync reset will
just continue and in case of fw fatal error no health reporting will be
done.
Fixes: 38b9f903f2 ("net/mlx5: Handle sync reset request event")
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Shay Drori <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 79a0e32b32 ]
The reclaim_pages_cmd() function sends a command to the firmware to
reclaim pages if the command interface is active.
A race condition can occur if the command interface goes down (e.g., due
to a PCI error) while the mlx5_cmd_do() call is in flight. In this
case, mlx5_cmd_do() will return an error. The original code would
propagate this error immediately, bypassing the software-based page
reclamation logic that is supposed to run when the command interface is
down.
Fix this by checking whether mlx5_cmd_do() returns -ENXIO, which mark
that command interface is down. If this is the case, fall through to
the software reclamation path. If the command failed for any another
reason, or finished successfully, return as before.
Fixes: b898ce7bcc ("net/mlx5: cmdif, Avoid skipping reclaim pages if FW is not accessible")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b1f0349bd6 ]
Stop polling on firmware response to command in polling mode if the
command interface got down. This situation can occur, for example, if a
firmware fatal error is detected during polling.
This change halts the polling process when the command interface goes
down, preventing unnecessary waits.
Fixes: b898ce7bcc ("net/mlx5: cmdif, Avoid skipping reclaim pages if FW is not accessible")
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Shay Drori <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 8169a6011c ]
The driver did not handle failure of `netdev_alloc_skb_ip_align()`.
If the allocation failed, dereferencing `skb->protocol` could lead to
a NULL pointer dereference.
This patch tries to allocate `skb`. If the allocation fails, it falls
back to the normal path.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Tested-on: D-Link DGE-550T Rev-A3
Signed-off-by: Yeounsu Moon <yyyynoom@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20250928190124.1156-1-yyyynoom@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f017156aea ]
In EC2 instances where the RSS hash key is not configurable, ethtool
shows bogus RSS hash key since ena_get_rxfh_key_size() unconditionally
returns ENA_HASH_KEY_SIZE.
Commit 6a4f7dc82d ("net: ena: rss: do not allocate key when not
supported") added proper handling for devices that don't support RSS
hash key configuration, but ena_get_rxfh_key_size() has been unchanged.
When the RSS hash key is not configurable, return 0 instead of
ENA_HASH_KEY_SIZE to clarify getting the value is not supported.
Tested on m5 instance families.
Without patch:
# ethtool -x ens5 | grep -A 1 "RSS hash key"
RSS hash key:
00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00
With patch:
# ethtool -x ens5 | grep -A 1 "RSS hash key"
RSS hash key:
Operation not supported
Fixes: 6a4f7dc82d ("net: ena: rss: do not allocate key when not supported")
Signed-off-by: Kohei Enju <enjuk@amazon.com>
Link: https://patch.msgid.link/20250929050247.51680-1-enjuk@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 8425161ac1 ]
The nfp_net_get_rxfh_key_size() function returns -EOPNOTSUPP when
devices don't support RSS, and callers treat the negative value as a
large positive value since the return type is u32.
Return 0 when devices don't support RSS, aligning with the ethtool
interface .get_rxfh_key_size() that requires returning 0 in such cases.
Fixes: 9ff304bfaf ("nfp: add support for reporting CRC32 hash function")
Signed-off-by: Kohei Enju <enjuk@amazon.com>
Link: https://patch.msgid.link/20250929054230.68120-1-enjuk@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b9bd25f47e ]
The mailbox receive path allocates coherent DMA memory with
dma_alloc_coherent(), but frees it with dmam_free_coherent().
This is incorrect since dmam_free_coherent() is only valid for
buffers allocated with dmam_alloc_coherent().
Fix the mismatch by using dma_free_coherent() instead of
dmam_free_coherent
Fixes: e54232da12 ("idpf: refactor idpf_recv_mb_msg")
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Madhu Chittim <madhu.chittim@intel.com>
Link: https://patch.msgid.link/20250925180212.415093-1-alok.a.tiwari@oracle.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 78d901897b ]
Move from 2*NUM_QUEUES dma_alloc_coherent() for DMA descriptor rings to
2 calls overall.
Issue is with how all queues share the same register for configuring the
upper 32-bits of Tx/Rx descriptor rings. Taking Tx, notice how TBQPH
does *not* depend on the queue index:
#define GEM_TBQP(hw_q) (0x0440 + ((hw_q) << 2))
#define GEM_TBQPH(hw_q) (0x04C8)
queue_writel(queue, TBQP, lower_32_bits(queue->tx_ring_dma));
#ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
if (bp->hw_dma_cap & HW_DMA_CAP_64B)
queue_writel(queue, TBQPH, upper_32_bits(queue->tx_ring_dma));
#endif
To maximise our chances of getting valid DMA addresses, we do a single
dma_alloc_coherent() across queues. This improves the odds because
alloc_pages() guarantees natural alignment. Other codepaths (IOMMU or
dev/arch dma_map_ops) don't give high enough guarantees
(even page-aligned isn't enough).
Two consideration:
- dma_alloc_coherent() gives us page alignment. Here we remove this
constraint meaning each queue's ring won't be page-aligned anymore.
- This can save some tiny amounts of memory. Fewer allocations means
(1) less overhead (constant cost per alloc) and (2) less wasted bytes
due to alignment constraints.
Example for (2): 4 queues, default ring size (512), 64-bit DMA
descriptors, 16K pages:
- Before: 8 allocs of 8K, each rounded to 16K => 64K wasted.
- After: 2 allocs of 32K => 0K wasted.
Fixes: 02c958dd34 ("net/macb: add TX multiqueue support for gem")
Reviewed-by: Sean Anderson <sean.anderson@linux.dev>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Tested-by: Nicolas Ferre <nicolas.ferre@microchip.com> # on sam9x75
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250923-macb-fixes-v6-4-772d655cdeb6@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 92d4256faf ]
The tx/rx ring size calculation is somewhat complex and partially hidden
behind a macro. Move that out of the {RX,TX}_RING_BYTES() macros and
macb_{alloc,free}_consistent() functions into neat separate functions.
In macb_free_consistent(), we drop the size variable and directly call
the size helpers in the arguments list. In macb_alloc_consistent(), we
keep the size variable that is used by netdev_dbg() calls.
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250923-macb-fixes-v6-3-772d655cdeb6@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Stable-dep-of: 78d901897b ("net: macb: single dma_alloc_coherent() for DMA descriptors")
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit fca3dc859b ]
The MACB driver acts as if TBQPH/RBQPH are configurable on a per queue
basis; this is a lie. A single register configures the upper 32 bits of
each DMA descriptor buffers for all queues.
Concrete actions:
- Drop GEM_TBQPH/GEM_RBQPH macros which have a queue index argument.
Only use MACB_TBQPH/MACB_RBQPH constants.
- Drop struct macb_queue->TBQPH/RBQPH fields.
- In macb_init_buffers(): do a single write to TBQPH and RBQPH for all
queues instead of a write per queue.
- In macb_tx_error_task(): drop the write to TBQPH.
- In macb_alloc_consistent(): if allocations give different upper
32-bits, fail. Previously, it would have lead to silent memory
corruption as queues would have used the upper 32 bits of the alloc
from queue 0 and their own low 32 bits.
- In macb_suspend(): if we use the tie off descriptor for suspend, do
the write once for all queues instead of once per queue.
Fixes: fff8019a08 ("net: macb: Add 64 bit addressing support for GEM")
Fixes: ae1f2a56d2 ("net: macb: Added support for many RX queues")
Reviewed-by: Sean Anderson <sean.anderson@linux.dev>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250923-macb-fixes-v6-2-772d655cdeb6@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c20edbacc0 ]
No idea what the current barrier position was meant for. At that point,
nothing is read from the descriptor, only the pointer to the actual one
is fetched.
The correct barrier usage here is after the generation check, so that
only the first qword is read if the descriptor is not yet ready and we
need to stop polling. Debatable on coherent DMA as the Rx descriptor
size is <= cacheline size, but anyway, the current barrier position
only makes the codegen worse.
Fixes: 3a8845af66 ("idpf: add RX splitq napi poll support")
Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Tested-by: Ramu R <ramu.r@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
This code calls kfree_rcu(new_node, rcu) and then dereferences "new_node"
and then dereferences it on the next line. Two lines later, we take
a mutex so I don't think this is an RCU safe region. Re-order it to do
the dereferences before queuing up the free.
Fixes: 68fbff68db ("octeontx2-pf: Add police action for TC flower")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/aNKCL1jKwK8GRJHh@stanley.mountain
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The LIBIE_AQ_STR macro() introduced by commit 5feaa7a07b ("libie: add
adminq helper for converting err to str") is used in order to generate
strings for printing human readable error codes. Its definition is missing
the separating underscore ('_') character which makes the resulting strings
difficult to read. Additionally, the string won't match the source code,
preventing search tools from working properly.
Add the missing underscore character, fixing the error string names.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Fixes: 5feaa7a07b ("libie: add adminq helper for converting err to str")
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://patch.msgid.link/20250923205657.846759-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>