The 'adjfine' method was first introduced in Linux v4.10 by commit
d8d263541913 ("ptp: Introduce a high resolution frequency adjustment
method.") and the 'adjfreq' method was finally removed in Linux v6.2.
Given that 'adjfine' has been supported since Linux v4.10, drop the
deprecated 'adjfreq' method completely to simplify the nvethernet
driver. This fixes build issues for 3rd party kernels that removed the
'adjfreq' method and are older than Linux v6.2.
Bug 3936429
Change-Id: I897bbd3ba92b11d8132c5762d2618b4e82ad44aa
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3128119
(cherry picked from commit 46af0c1d24)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3129149
Reviewed-by: Brad Griffis <bgriffis@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
For Linux v6.9, the structure 'ethtool_eee' was replaced by
'ethtool_keee' and this new structure uses bitmaps for the structure
fields 'supported', 'advertised' and 'lp_advertised' as opposed to
'u32' types.
Add a test to conftest to detect the presence of the new 'ethtool_keee'
structure and update the nvethernet driver to use the various
linkmode_xxx helper functions for accessing the bitmap structure fields.
Bug 4471899
Change-Id: I275becfa665a36ed90e8bc227eaa4122825faa6a
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3110494
(cherry picked from commit 72761b8276)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3111548
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Existing implementation uses a single page_pool
for all Rx DMA channels. As by default all irqs
are on CPU 0, this does not cause any issue.
Routing an irq to some other CPU causes race
conditon for page allocation from page_pool which
leads to random memory corruption.
[ 158.416637] Call trace:
[ 158.416644] arm_lpae_map_pages+0xb4/0x1e0
[ 158.416649] arm_smmu_map_pages+0x8c/0x160
[ 158.416661] __iommu_map+0xf8/0x2b0
[ 158.416677] iommu_map_atomic+0x58/0xb0
[ 158.416683] __iommu_dma_map+0xac/0x150
[ 158.416687] iommu_dma_map_page+0xf4/0x220
[ 158.416690] dma_map_page_attrs+0x1e8/0x260
[ 158.416727] page_pool_dma_map+0x48/0xd0
[ 158.416750] __page_pool_alloc_pages_slow+0xc4/0x390
[ 158.416757] page_pool_alloc_pages+0x64/0x90
[ 158.416762] ether_padctrl_mii_rx_pins+0x164/0x9d0 [nvethernet]
[ 158.416807] ether_padctrl_mii_rx_pins+0x478/0x9d0 [nvethernet]
[ 158.416822] osi_process_rx_completions+0x284/0x4d0 [nvethernet]
[ 158.416832] 0xffffa26218b8f71c
[ 158.416855] __napi_poll+0x48/0x230
[ 158.416871] net_rx_action+0xf4/0x290
[ 158.416875] __do_softirq+0x130/0x3e8
[ 158.416889] __irq_exit_rcu+0xe8/0x110
This change creates a page_pool per Rx DMA channel.
This ensures there is no race conditon for page
alloc/dealloc from each DMA napi context.
Bug 4541158
Change-Id: I12668ee7d824fd30d54a874bbbdf190d02943478
Signed-off-by: Aniruddha Paul <anpaul@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3093534
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
If the userspace service 'irqbalance' is installed then the nvethernet
driver crashes when there is network activity. To avoid this crash set
the IRQF_NOBALANCING flash for the VM interrupts. No performance
degradation is observed when running iperf3 with a 1Gbps link.
Long-term the nvethernet driver still needs to be fixed to allow IRQ
balancing.
Bug 4293378
Bug 4541158
Change-Id: I0aa4ee28e36c7d273f14ff043544e72d3e988bd3
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3087525
Reviewed-by: Brad Griffis <bgriffis@nvidia.com>
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Add tests to conftest for detecting if the mii_bus structure has the
read_c45 and write_c45 function pointers and use the definitions
generated by conftest in the nvethernet driver.
This fixes support for nvethernet in 3rd party Linux kernels that have
backported the mii_bus structure changes to their kernel that have a
kernel version prior to Linux v6.3.
Bug 4014315
Change-Id: I5ae98fc5077337286921da6e9347df9781565a70
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3018935
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Split the type and declaration of page_pool header for Linux6.6
based on below change from kernel 6.6:
---
commit a9ca9f9ceff382b58b488248f0c0da9e157f5d06
Author: Yunsheng Lin <linyunsheng@huawei.com>
Date: Fri Aug 4 20:05:24 2023 +0200
page_pool: split types and declarations from page_pool.h
---
Bug 4346767
Change-Id: I0217da34c67d952c97c6812c2760bed2fb3df3e0
Signed-off-by: Laxman Dewangan <ldewangan@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3008090
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Issue:
The L2 Orin bridge use case is encountered
an issue on AV+L NDAS builds for VLAN ping
from HOST1 to HOST2 over L2 Orin bridge.
As HOST1/HOST2 VLAN packets are dropped
by EQOS/MGBE MAC due to VLAN VID feature enable.
Fix:
To support the L2 Orin bridge VLAN
HOST1 to HOST2 data usecase, disable
the VLAN VID filters in the nvethernet driver.
Bug 4230250
Change-Id: I02514fcc15b7764fe93118a9f90c16716368c73c
Signed-off-by: Mohan Thadikamalla <mohant@nvidia.com>
(cherry picked from commit 86cb37f933846a012ba6d8cdac275a2edda0aa42)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3003075
Reviewed-by: Bhadram Varka <vbhadram@nvidia.com>
Reviewed-by: Srinivas Ramachandran <srinivasra@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
In Linux v6.4, commit
3948b05950fd ("net: introduce a config option to tweak MAX_SKB_FRAGS")
added a kernel config option for MAX_SKB_FRAGS and change the type of
the definition to integer. This causes the nvethernet driver build to
fail and the following errors are seen ...
drivers/net/ethernet/nvidia/nvethernet/ether_linux.c:5949:33: error:
format ‘%ld’ expects argument of type ‘long int’, but argument 4
has type ‘nveu32_t’ {aka ‘unsigned int’} [-Werror=format=]
"invalid tx-frames, must be inrange %d to %ld",
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/nvidia/nvethernet/ethtool.c:950:28: error: format
‘%ld’ expects argument of type ‘long int’, but argument 4 has
type ‘nveu32_t’ {aka ‘unsigned int’} [-Werror=format=]
"invalid tx-frames, must be in the range of"
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Fix this by changing the type in the respective prints to integer. Note
that his is also to make this change for kernel prior to Linux v6.4.
Bug 4183168
Bug 4221847
Change-Id: Ibd9160ad08e65e0b85cae1cf4903075c31a1cd77
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2989002
Reviewed-by: Shanker Donthineni <sdonthineni@nvidia.com>
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Instead of relying on kernel version to determine if functions or
specific versions of functions are present in the kernel, add compile
time tests to the conftest.sh script to determine this at compile time
for the kernel being used. This is beneficial for working with 3rd party
Linux kernels that may have back-ported upstream changes into their
kernel and so the kernel version checks do not work.
Bug 4119327
Change-Id: I79e701940ca70ca4d66500c75b5992f9d92b54b0
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2985744
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Issue:
L4T platforms on boot comes up with MTU as 1500.
However on changing the MTU to jumbo, the final
MTU which gets set accounts the MACSEC ICV tag
len and reduces the MTU by 34 bytes even though
the platform does not support MACSEC.
Fix:
Assign macsec allocated private data to NULL and
use the NULL check in ndo mtu change ioctl call
to reject MTU reduction if macsec is not supported.
Bug 4137597
Change-Id: Ica3003b54762dd5d240c9b0c5c012049b4e503b2
Signed-off-by: Sushil Kumar Singh <sushilkumars@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2928569
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-by: Bibek Basu <bbasu@nvidia.com>
Tested-by: Revanth Kumar Uppala <ruppala@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
ether_get_tx_ts, pdata->tx_ts_ref_cnt is an atomic variable
that is used as a mutex in this function and should not return 0
if the function fails to acquire the mutex.
The workqueue is scheduled when the return value of the
judgment function is <0. If one CPU core execution softirq
is running in ether_get_tx_ts function and another CPU core
softirq also calls ether_get_tx_ts function to get hardware
timestamp, then the acquisition of mutex fails and return 0.
After returning 0, the workqueue is not scheduled.
The timestamp cannot be obtained in time. So ether_get_tx_ts
should returns -1 on failure to acquire the mutex.
Bug 4150416
Change-Id: Icffd409b349d8bb8dbf5a483124b3bd3d7ef6cc8
Signed-off-by: Bhadram Varka <vbhadram@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2934350
Reviewed-by: Bitan Biswas <bbiswas@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
If DEBUGFS is not enabled, then the following warning is observed ...
drivers/net/ethernet/nvidia/nvethernet/sysfs.c:3321:6:
error: unused variable [-Werror=unused-variable]
3321 | int ret = 0;
| ^~~
Fix this by ensuring that 'ret' is only defined if the DEBUGFS is
enabled.
Bug 4190165
Change-Id: I0182874592f20abe8d53b7bfdcfd099518b23f9c
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2938450
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Compiling the nvethernet driver without the ccflag CONFIG_TEGRA_NVPPS
defined generates the following warning ...
drivers/net/ethernet/nvidia/nvethernet/ether_linux.c:2936:5:
warning: "CONFIG_TEGRA_NVPPS" is not defined, evaluates to 0 [-Wundef]
2936 | #if CONFIG_TEGRA_NVPPS
| ^~~~~~~~~~~~~~~~~~
Fix this by using '#ifdef' to determine if CONFIG_TEGRA_NVPPS is
defined instead of '#if'.
Bug 4190030
Change-Id: Id1523de6f9ef9f4c72669584efb77ffd9ddf20f3
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2938375
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Building the nvethernet driver generates the following warning ...
drivers/net/ethernet/nvidia/nvethernet/ether_linux.c:6730:1: warning:
the frame size of 3664 bytes is larger than 2048 bytes
[-Wframe-larger-than=]
For now allow the frame size to be 4096 bytes to resolve the warning
while a proper fix is implemented.
Bug 4190030
Change-Id: I5460ec04c80458174c2c51dd0058b7cc191fe808
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2934698
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Issue:
MGBE0 ASID is getting programmed for all the MAC instances
which inturn causing the data transfer failures.Macsec is not enable
and hence instance_id is not getting updated to the corresponding mac
instance number and hence all mac instance_id's are initialized with 0.
Since ASID programming is based on mac instance_id, all MACs are
programmed with MGBE0 MAC instance id.
Fix:
Moved reading of instance_id from DT out of macsec_probe and hence the
instance_id gets updated irrespective of MAC sec enable/disable.
Bug 4124937
Change-Id: I500ed8d3db402995488260af99ad190217ff6dd2
Signed-off-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2925252
Reviewed-by: Narayan Reddy <narayanr@nvidia.com>
Reviewed-by: Laxman Dewangan <ldewangan@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Commit db1a63aed89c ("net: phy: Remove fallback to old C45 method")
removes a fallback C45 method in the MDIO bus driver. This breaks
nvethernet support for Linux v6.3 and the following errors are observed.
failed to register MDIO bus (nvethernet_mdio_bus)
net eth0: failed to register MDIO bus
Fix this by adding the necessary read/write_c45 callbacks in the
nvethernet driver. Note these callbacks are only supported for Linux
v6.3+ kernels.
Bug 4014315
Change-Id: Ia6ab753941db0515799657da8522f32996d0852a
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2924619
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-by: Narayan Reddy <narayanr@nvidia.com>
The nvethernet driver has a dependency on the NVPPS driver which in
turn has a dependency on the GTE driver. The GTE driver is being
transitioned to the new upstream HTE driver and while this transition is
on-going, allow the nvethernet driver to be distributed and built
without NVPPS and GTE.
Note that by default the dependency on the NVPPS driver is enabled by
setting CONFIG_TEGRA_NVPPS to 'y'. Only if the user sets
CONFIG_TEGRA_NVPPS to 'n' will it be disabled.
Bug 3918941
Bug 3961133
Change-Id: I019e5ebfddfcb158b0a684d2eacc404231577166
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2913273
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-by: Bhadram Varka <vbhadram@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Issue:
Observed below compilation issues
on the nvethernet driver for DRIVE Linux safety
builds as some macros/functions are defined with
"OSI_STRIPPED_LIB" protection but they are used
directly without any protection.
1./automotive-dev-main-20230516T005516255/out/embedded-linux-generic-release-safety/nvidia/kernel-rt_patches-nvidia-oot/nvidia-oot/drivers/net/ethernet/nvidia/nvethernet/sysfs.c:2641:33: error: ‘OSI_INSTANCE_ID_EQOS’ undeclared (first use in this function)
2641 | if (osi_core->instance_id == OSI_INSTANCE_ID_EQOS)
| ^~~~~~~~~~~~~~~~~~~~
2./automotive-dev-main-20230516T005516255/out/embedded-linux-generic-release-safety/nvidia/kernel-rt_patches-nvidia-oot/nvidia-oot/drivers/net/ethernet/nvidia/nvethernet/sysfs.c:2733:3: error: ‘dev_attr_hsi_enable’ undeclared here (not in a function)
2733 | &dev_attr_hsi_enable.attr,
| ^~~~~~~~~~~~~~~~~~~
Fix:
Add OSI_STRIPPED_LIB macro protection
for required functions.
Bug 3959085
Change-Id: Ifef1c72b5ddedc4d550280eace305bbade290bbf
Signed-off-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2905357
Reviewed-by: Laxman Dewangan <ldewangan@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Issue: skb_vlan_tag_get gives only the VLAN_ID for
the earlier kernel versions (K4.9) and hence added
a logic to take care of adding the VLAN priority
explicitly and hence the VLAN priority getting
modified in a newer kernel version where VLAN PRIO is
already included
Fix: remove the logic of explicit adding of VLAN
priority.
Bug 3788862
Bug 4088361
Change-Id: I0ac9ebfe21d3e696ac1b0f6ba540c010928f775f
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2806468
(cherry picked from commit 799b17fc35f3bc3ec75ea93452c03bd52ed28527)
Signed-off-by: Narayan Reddy <narayanr@nvidia.com>
Signed-off-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2895043
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Issue:
During eth0 down case, the link change gets called twice,
1. due to ndo close
2. due to interrupt generated by phy.
The ether_adjust_link callback also configures the EEE settings
and calls phy_init_eee (a phy framework API) for the same.
Here a race condition occur, when ether_close sets phydev to NULL
before the phy_init_eee gets called (from ether_conf_eee) which tries
to deference the phydev causing kernel panic.
[35779.541305] Call trace:
[35779.544018] phy_init_eee+0x24/0x290
[35779.547455] ether_conf_eee+0x110/0x1310 [nvethernet]
[35779.552525] ether_conf_eee+0x4e0/0x1310 [nvethernet]
[35779.557840] phy_link_change+0x40/0x90
[35779.561603] phy_state_machine+0x190/0x240
[35779.565643] process_one_work+0x1c4/0x4d0
[35779.569396] worker_thread+0x54/0x430
[35779.573072] kthread+0x148/0x180
[35779.576586] ret_from_fork+0x10/0x34
[35779.580069] Code: a90153f3 d5384113 a9025bf5 12001c36 (f941b801)
[35779.586195] ---[ end trace 2ea5b6492f36b0cb ]---
[35779.590737] Kernel panic - not syncing: Oops: Fatal exception
Fix:
Add a NULL check and return -ENODEVICE in EEE configuration
Bug 3877272
Bug 3920560
Bug 3865666
Bug 4088361
Change-Id: I2b26c6eeabb7a109b9d14b3bd2a035f5c3b3c6fa
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2854404
(cherry picked from commit 40606dc247eacf2a8987687777874575d9037058)
Signed-off-by: Sushil Kumar Singh <sushilkumars@nvidia.com>
Signed-off-by: Revanth Kumar Uppala <ruppala@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2855237
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Issue: If the supplicant is killed for some reason Data would flow
plain on that interface, this needs to be avoided
Fix: Update bypass LUT such that if the frames from the VF(on which
supplicant is launched) is received on MACSEC either authenticate the
same or drop. Along with this handles below items as well. All the VFs MACIDs
are obtained in OSI to update the bypass LUTs to decide on which VF frames
to be authenticated and which VF frames needs to be bypassed.
1. Remove osi_macsec_en API and have single API to init and deinit
2. Remove explicit command from supplicant to set control port and
set protected frames. Handle the same in osi_macsec_init
Bug 3984665
Change-Id: I8bc8aa95d1e21e99e992b471fb70ed58073163f7
Signed-off-by: Sanath Kumar Gampa <sgampa@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2878515
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Upstream commit 99d5fe9c7f3d ("net: mdio: Remove support for building C45
muxed addresses") removed the definition MII_ADDR_C45 breaking the build
for the nvethernet driver for Linux v6.3. Update the driver to use the
OSI_MII_ADDR_C45 instead.
Bug 4014315
Bug 4075345
Change-Id: I98f8d10ce1a093d458d293b286a4e5a544d48d04
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2889876
Reviewed-by: Rohit Khanna <rokhanna@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
