| Age | Commit message (Collapse) | Author |
|---|
|
After addition of HW managed counters and implementation drop
in flow steering logic, the code in driver which checks syndrome
is not reachable anymore.
Let's delete it.
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Fill integrity, replay and bad trailer counters.
As an example, after simulating replay window attack with 5 packets:
[leonro@c ~]$ grep XfrmInStateSeqError /proc/net/xfrm_stat
XfrmInStateSeqError 5
[leonro@c ~]$ sudo ip -s x s
<...>
stats:
replay-window 0 replay 5 failed 0
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
Iterate over all SAs in order to fill global IPsec statistics.
Acked-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
In order to allow drivers to fill all statistics, change the name
of xdo_dev_state_update_curlft to be xdo_dev_state_update_stats.
Acked-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
|
|
commit 1c870c63d7d2 ("net: fill in MODULE_DESCRIPTION()s for ocelot")
got a suggestion from Vladimir Oltean after it had landed in net-next.
Rewrite the module description according to Vladimir's suggestion.
Fixes: 1c870c63d7d2 ("net: fill in MODULE_DESCRIPTION()s for ocelot")
Suggested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Similar chunk of code is used in tsnep_rx_poll_zc() and
tsnep_rx_reopen_xsk() to maintain the RX XDP_RING_NEED_WAKEUP flag.
Consolidate the code to common helper function.
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Enable previously excluded xdp feature flag for NFD3 devices. This
feature flag is required in order to bind nfp interfaces to an xdp
socket and the nfp driver does in fact support the feature.
Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Cc: stable@vger.kernel.org # 6.3+
Signed-off-by: James Hershaw <james.hershaw@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When physical ports are reset (either through link failure or manually
toggled down and up again) that are slaved to a Linux bond with a tunnel
endpoint IP address on the bond device, not all tunnel packets arriving
on the bond port are decapped as expected.
The bond dev assigns the same MAC address to itself and each of its
slaves. When toggling a slave device, the same MAC address is therefore
offloaded to the NFP multiple times with different indexes.
The issue only occurs when re-adding the shared mac. The
nfp_tunnel_add_shared_mac() function has a conditional check early on
that checks if a mac entry already exists and if that mac entry is
global: (entry && nfp_tunnel_is_mac_idx_global(entry->index)). In the
case of a bonded device (For example br-ex), the mac index is obtained,
and no new index is assigned.
We therefore modify the conditional in nfp_tunnel_add_shared_mac() to
check if the port belongs to the LAG along with the existing checks to
prevent a new global mac index from being re-assigned to the slave port.
Fixes: 20cce8865098 ("nfp: flower: enable MAC address sharing for offloadable devs")
CC: stable@vger.kernel.org # 5.1+
Signed-off-by: Daniel de Villiers <daniel.devilliers@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The 1st and 2nd expansion BAR configuration registers are configured,
when the driver starts up, in variables 'barcfg_msix_general' and
'barcfg_msix_xpb', respectively. The 'LengthSelect' field is ORed in
from bit 0, which is incorrect. The 'LengthSelect' field should
start from bit 27.
This has largely gone un-noticed because
NFP_PCIE_BAR_PCIE2CPP_LengthSelect_32BIT happens to be 0.
Fixes: 4cb584e0ee7d ("nfp: add CPP access core")
Cc: stable@vger.kernel.org # 4.11+
Signed-off-by: Daniel Basilio <daniel.basilio@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We need the tty/serial fixes in here as well.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
This adds support for the RTL8126A found on Asus z790 Maximus Formula.
It was successfully tested w/o the firmware at 1000Mbps. Firmware file
has been provided by Realtek and submitted to linux-firmware.
2.5G and 5G modes are untested.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When qmem_alloc and pfvf->hw_ops->sq_aq_init fails, sq->sg should be
freed to prevent memleak.
Fixes: c9c12d339d93 ("octeontx2-pf: Add support for PTP clock")
Signed-off-by: Zhipeng Lu <alexious@zju.edu.cn>
Acked-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For XDP_TX action xdp_buff is converted to xdp_frame. The conversion is
done by xdp_convert_buff_to_frame(). The memory type of the resulting
xdp_frame depends on the memory type of the xdp_buff. For page pool
based xdp_buff it produces xdp_frame with memory type
MEM_TYPE_PAGE_POOL. For zero copy XSK pool based xdp_buff it produces
xdp_frame with memory type MEM_TYPE_PAGE_ORDER0.
tsnep_xdp_xmit_back() is not prepared for that and uses always the page
pool buffer type TSNEP_TX_TYPE_XDP_TX. This leads to invalid mappings
and the transmission of undefined data.
Improve tsnep_xdp_xmit_back() to use the generic buffer type
TSNEP_TX_TYPE_XDP_NDO for zero copy XDP_TX.
Fixes: 3fc2333933fd ("tsnep: Add XDP socket zero-copy RX support")
Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We don't have to store the EEE modes to be advertised in the driver,
phylib does this for us and stores it in phydev->advertising_eee.
phylib also takes care of properly handling the EEE advertisement.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://lore.kernel.org/r/27c336a8-ea47-483d-815b-02c45ae41da2@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Function aq_ring_hwts_rx_alloc() maps extra AQ_CFG_RXDS_DEF bytes
for PTP HWTS ring but then generic aq_ring_free() does not take this
into account.
Create and use a specific function to free HWTS ring to fix this
issue.
Trace:
[ 215.351607] ------------[ cut here ]------------
[ 215.351612] DMA-API: atlantic 0000:4b:00.0: device driver frees DMA memory with different size [device address=0x00000000fbdd0000] [map size=34816 bytes] [unmap size=32768 bytes]
[ 215.351635] WARNING: CPU: 33 PID: 10759 at kernel/dma/debug.c:988 check_unmap+0xa6f/0x2360
...
[ 215.581176] Call Trace:
[ 215.583632] <TASK>
[ 215.585745] ? show_trace_log_lvl+0x1c4/0x2df
[ 215.590114] ? show_trace_log_lvl+0x1c4/0x2df
[ 215.594497] ? debug_dma_free_coherent+0x196/0x210
[ 215.599305] ? check_unmap+0xa6f/0x2360
[ 215.603147] ? __warn+0xca/0x1d0
[ 215.606391] ? check_unmap+0xa6f/0x2360
[ 215.610237] ? report_bug+0x1ef/0x370
[ 215.613921] ? handle_bug+0x3c/0x70
[ 215.617423] ? exc_invalid_op+0x14/0x50
[ 215.621269] ? asm_exc_invalid_op+0x16/0x20
[ 215.625480] ? check_unmap+0xa6f/0x2360
[ 215.629331] ? mark_lock.part.0+0xca/0xa40
[ 215.633445] debug_dma_free_coherent+0x196/0x210
[ 215.638079] ? __pfx_debug_dma_free_coherent+0x10/0x10
[ 215.643242] ? slab_free_freelist_hook+0x11d/0x1d0
[ 215.648060] dma_free_attrs+0x6d/0x130
[ 215.651834] aq_ring_free+0x193/0x290 [atlantic]
[ 215.656487] aq_ptp_ring_free+0x67/0x110 [atlantic]
...
[ 216.127540] ---[ end trace 6467e5964dd2640b ]---
[ 216.132160] DMA-API: Mapped at:
[ 216.132162] debug_dma_alloc_coherent+0x66/0x2f0
[ 216.132165] dma_alloc_attrs+0xf5/0x1b0
[ 216.132168] aq_ring_hwts_rx_alloc+0x150/0x1f0 [atlantic]
[ 216.132193] aq_ptp_ring_alloc+0x1bb/0x540 [atlantic]
[ 216.132213] aq_nic_init+0x4a1/0x760 [atlantic]
Fixes: 94ad94558b0f ("net: aquantia: add PTP rings infrastructure")
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20240201094752.883026-1-ivecera@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Copy paste issue left a comment for this structure that has nothing
to do with FW alignment; remove the comment.
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
|
|
HW incorrectly reports EIPE errors on encapsulated packets
with L2 padding inside inner packet. HW shows outer UDP/IPV4
packet checksum errors as part of the EIPE flags of the
Rx descriptor. These are reported only if checksum offload
is enabled and L3/L4 parsed flag is valid in Rx descriptor.
When that error is reported by HW, we don't act on it
instead of incrementing main Rx errors statistic as it
would normally happen.
Add a new statistic to count these errors since we still want
to print them.
Signed-off-by: Aniruddha Paul <aniruddha.paul@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Jan Glaza <jan.glaza@intel.com>
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
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>
|
|
Currently, XSK control path in ice driver calls directly
ice_vsi_cfg_txq() whereas we have ice_vsi_cfg_single_txq() for that
purpose. Use the latter from XSK side and make ice_vsi_cfg_txq() static.
ice_vsi_cfg_txq() resides in ice_base.c and is rather big, so to reduce
the code churn let us move the callers of it from ice_lib.c to
ice_base.c.
This change puts ice_qp_ena() on nice diet due to the checks and
operations that ice_vsi_cfg_single_{r,t}xq() do internally.
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-182 (-182)
Function old new delta
ice_xsk_pool_setup 2165 1983 -182
Total: Before=472597, After=472415, chg -0.04%
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
|
|
Currently, XSK control path in ice driver calls directly
ice_vsi_cfg_rxq() whereas we have ice_vsi_cfg_single_rxq() for that
purpose. Use the latter from XSK side and make ice_vsi_cfg_rxq() static.
ice_vsi_cfg_rxq() resides in ice_base.c and is rather big, so to reduce
the code churn let us move two callers of it from ice_lib.c to
ice_base.c.
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
|
|
PCI device IDs of RVU device IDs are configurable and
RVU PF0's (ie AF's) are currently assumed as VFs that
identify loopback functionality ie LBKVFs. But in some
cases these VFs can be setup for different functionality.
Hence remove assumptions that AF's VFs are always LBK VFs
by renaming 'is_afvf' as 'is_lbkvf' explicitly and also
identify LBK VF using PCI dev ID. Similar change is done
for other VF types.
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In current driver 64 BPIDs are reserved for LBK interfaces.
These bpids are 1-to-1 mapped to LBK interface channel numbers.
In some usecases one LBK interface required more than one
bpids and in some case they may not require at all.
These usescase can't be address with the current implementation
as it always reserves only one bpid per LBK channel.
This patch addresses this issue by creating free bpid pool from these
64 bpids instead of 1-to-1 mapping to the lbk channel.
Now based on usecase LBK interface can request a bpid using (bp_enable()).
This patch also reduces the number of bpids for cgx interfaces to 8
and adds proper error code
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Commit 56e58d6c8a56 ("net: stmmac: Implement Safety Features in
XGMAC core") checks and reports safety errors, but leaves the
Data Path Parity Errors for each channel in DMA unhandled at all, lead to
a storm of interrupt.
Fix it by checking and clearing the DMA_DPP_Interrupt_Status register.
Fixes: 56e58d6c8a56 ("net: stmmac: Implement Safety Features in XGMAC core")
Signed-off-by: Furong Xu <0x1207@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Cross-merge networking fixes after downstream PR.
No conflicts or adjacent changes.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
In the arm random config file, kconfig option 'CONFIG_AEABI' is
disabled which results in adding the compiler flag '-mabi=apcs-gnu'.
This causes the compiler to add padding in virtchnl2_ptype
structure to align it to 8 bytes, resulting in the following
size check failure:
include/linux/build_bug.h:78:41: error: static assertion failed: "(6) == sizeof(struct virtchnl2_ptype)"
78 | #define __static_assert(expr, msg, ...) _Static_assert(expr, msg)
| ^~~~~~~~~~~~~~
include/linux/build_bug.h:77:34: note: in expansion of macro '__static_assert'
77 | #define static_assert(expr, ...) __static_assert(expr, ##__VA_ARGS__, #expr)
| ^~~~~~~~~~~~~~~
drivers/net/ethernet/intel/idpf/virtchnl2.h:26:9: note: in expansion of macro 'static_assert'
26 | static_assert((n) == sizeof(struct X))
| ^~~~~~~~~~~~~
drivers/net/ethernet/intel/idpf/virtchnl2.h:982:1: note: in expansion of macro 'VIRTCHNL2_CHECK_STRUCT_LEN'
982 | VIRTCHNL2_CHECK_STRUCT_LEN(6, virtchnl2_ptype);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
Avoid the compiler padding by using "__packed" structure
attribute for the virtchnl2_ptype struct. Also align the
structure by using "__aligned(2)" for better code optimization.
Fixes: 0d7502a9b4a7 ("virtchnl: add virtchnl version 2 ops")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202312220250.ufEm8doQ-lkp@intel.com
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Pavan Kumar Linga <pavan.kumar.linga@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20240131222241.2087516-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Fill-up the lock status error value properly.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Acked-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Pass additional argunent status_error over lock_status_get()
so drivers can fill it up. In case they do, expose the value over
previously introduced attribute to user. Do it only in case the
current lock_status is either "unlocked" or "holdover".
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Acked-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
XDP queues are created/destroyed when a XDP program
is attached/detached. In current driver xdp_queues are not
getting destroyed on program exit due to incorrect xdp_queue
and tot_tx_queue count values.
This patch fixes the issue by setting tot_tx_queue and xdp_queue
count to correct values. It also fixes xdp.data_hard_start address.
Fixes: 06059a1a9a4a ("octeontx2-pf: Add XDP support to netdev PF")
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Link: https://lore.kernel.org/r/20240130120610.16673-1-gakula@marvell.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
This patch reduces some of the lines by removing newlines
where more variables or print strings can be pushed back
to the previous line while still adhering to the styling
guidelines.
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Fail queue size calculation when the device returns maximum
TX/RX queue sizes that are smaller than the allowed minimum.
Signed-off-by: Osama Abboud <osamaabb@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
The netif_* functions are used by the driver to log events into the
kernel ring (dmesg) similar to the netdev_* ones. Unlike the latter,
the netif_* function family allow the user to choose what events get
logged using ethtool:
sudo ethtool -s [interface] msglvl [msg_type] on
By default the events which get logged are slow-path related and aren't
printed often (e.g. interface up related prints). This patch removes the
NETIF_MSG_TX_DONE type (called every TX completion polling) from the
defaults and adds NETIF_MSG_IFDOWN instead as it makes more sensible
defaults.
This patch also transforms ena_down() print from netif_info into
netif_dbg (same as the analogue print in ena_up()) as it suits it
better.
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Move skb_tx_timestamp() closer to the actual time the driver sends the
packets to the device.
Signed-off-by: Osama Abboud <osamaabb@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
The function responsible for polling TX completions might not receive
the CPU resources it needs due to higher priority tasks running on the
requested core.
The driver might not be able to recognize such cases, but it can use its
state to suspect that they happened. If both conditions are met:
- napi hasn't been executed more than the TX completion timeout value
- napi is scheduled (meaning that we've received an interrupt)
Then it's more likely that the napi handler isn't scheduled because of
an overloaded CPU.
It was decided that for this case, the driver would wait twice as long
as the regular timeout before scheduling a reset.
The driver uses ENA_REGS_RESET_SUSPECTED_POLL_STARVATION reset reason to
indicate this case to the device.
This patch also adds more information to the ena_tx_timeout() callback.
This function is called by the kernel when it detects that a specific TX
queue has been closed for too long.
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
The print was re-worded to a more informative one.
Signed-off-by: Shahar Itzko <itzko@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
The functionality was added to allow the drivers to create an
SQ and CQ of different sizes.
When the RX/TX SQ and CQ have the same size, such update isn't
necessary as the device can safely assume it doesn't override
unprocessed completions. However, if the SQ is larger than the CQ,
the device might "have" more completions it wants to update about
than there's room in the CQ.
There's no support for different SQ and CQ sizes, therefore,
removing the API and its usage.
'____cacheline_aligned' compiler attribute was added to
'struct ena_com_io_cq' to ensure that the removal of the
'cq_head_db_reg' field doesn't change the cache-line layout
of this struct.
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Dynamic Interrupt Moderation (DIM) is a technique
designed to balance the need for timely data processing
with the desire to minimize CPU overhead.
Instead of generating an interrupt for every received
packet, the system can dynamically adjust the rate at
which interrupts are generated based on the incoming
traffic patterns.
Enabling DIM by default to improve the user experience.
DIM can be turned on/off through ethtool:
`ethtool -C <interface> adaptive-rx <on/off>`
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Osama Abboud <osamaabb@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
A few changes for better readability and style
1. Adding / Removing newlines
2. Removing an unnecessary and confusing comment
3. Using an existing variable rather than re-checking a field
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Remove io_sq->header_addr field because it is no longer
in use.
LLQ was updated to support a bounce buffer so there is
no need in saving the header address of the sq.
Signed-off-by: Nati Koler <nkoler@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Existing MANA design assigns IRQ to every CPU, including sibling
hyper-threads. This may cause multiple IRQs to be active simultaneously
in the same core and may reduce the network performance.
Improve the performance by assigning IRQ to non sibling CPUs in local
NUMA node. The performance improvement we are getting using ntttcp with
following patch is around 15 percent against existing design and
approximately 11 percent, when trying to assign one IRQ in each core
across NUMA nodes, if enough cores are present.
The change will improve the performance for the system
with high number of CPU, where number of CPUs in a node is more than
64 CPUs. Nodes with 64 CPUs or less than 64 CPUs will not be affected
by this change.
The performance study was done using ntttcp tool in Azure.
The node had 2 nodes with 32 cores each, total 128 vCPU and number of channels
were 32 for 32 RX rings.
The below table shows a comparison between existing design and new
design:
IRQ node-num core-num CPU performance(%)
1 0 | 0 0 | 0 0 | 0-1 0
2 0 | 0 0 | 1 1 | 2-3 3
3 0 | 0 1 | 2 2 | 4-5 10
4 0 | 0 1 | 3 3 | 6-7 15
5 0 | 0 2 | 4 4 | 8-9 15
...
...
25 0 | 0 12| 24 24| 48-49 12
...
32 0 | 0 15| 31 31| 62-63 12
33 0 | 0 16| 0 32| 0-1 10
...
64 0 | 0 31| 31 63| 62-63 0
Signed-off-by: Souradeep Chakrabarti <schakrabarti@linux.microsoft.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Souradeep investigated that the driver performs faster if IRQs are
spread on CPUs with the following heuristics:
1. No more than one IRQ per CPU, if possible;
2. NUMA locality is the second priority;
3. Sibling dislocality is the last priority.
Let's consider this topology:
Node 0 1
Core 0 1 2 3
CPU 0 1 2 3 4 5 6 7
The most performant IRQ distribution based on the above topology
and heuristics may look like this:
IRQ Nodes Cores CPUs
0 1 0 0-1
1 1 1 2-3
2 1 0 0-1
3 1 1 2-3
4 2 2 4-5
5 2 3 6-7
6 2 2 4-5
7 2 3 6-7
The irq_setup() routine introduced in this patch leverages the
for_each_numa_hop_mask() iterator and assigns IRQs to sibling groups
as described above.
According to [1], for NUMA-aware but sibling-ignorant IRQ distribution
based on cpumask_local_spread() performance test results look like this:
./ntttcp -r -m 16
NTTTCP for Linux 1.4.0
---------------------------------------------------------
08:05:20 INFO: 17 threads created
08:05:28 INFO: Network activity progressing...
08:06:28 INFO: Test run completed.
08:06:28 INFO: Test cycle finished.
08:06:28 INFO: ##### Totals: #####
08:06:28 INFO: test duration :60.00 seconds
08:06:28 INFO: total bytes :630292053310
08:06:28 INFO: throughput :84.04Gbps
08:06:28 INFO: retrans segs :4
08:06:28 INFO: cpu cores :192
08:06:28 INFO: cpu speed :3799.725MHz
08:06:28 INFO: user :0.05%
08:06:28 INFO: system :1.60%
08:06:28 INFO: idle :96.41%
08:06:28 INFO: iowait :0.00%
08:06:28 INFO: softirq :1.94%
08:06:28 INFO: cycles/byte :2.50
08:06:28 INFO: cpu busy (all) :534.41%
For NUMA- and sibling-aware IRQ distribution, the same test works
15% faster:
./ntttcp -r -m 16
NTTTCP for Linux 1.4.0
---------------------------------------------------------
08:08:51 INFO: 17 threads created
08:08:56 INFO: Network activity progressing...
08:09:56 INFO: Test run completed.
08:09:56 INFO: Test cycle finished.
08:09:56 INFO: ##### Totals: #####
08:09:56 INFO: test duration :60.00 seconds
08:09:56 INFO: total bytes :741966608384
08:09:56 INFO: throughput :98.93Gbps
08:09:56 INFO: retrans segs :6
08:09:56 INFO: cpu cores :192
08:09:56 INFO: cpu speed :3799.791MHz
08:09:56 INFO: user :0.06%
08:09:56 INFO: system :1.81%
08:09:56 INFO: idle :96.18%
08:09:56 INFO: iowait :0.00%
08:09:56 INFO: softirq :1.95%
08:09:56 INFO: cycles/byte :2.25
08:09:56 INFO: cpu busy (all) :569.22%
[1] https://lore.kernel.org/all/20231211063726.GA4977@linuxonhyperv3.guj3yctzbm1etfxqx2vob5hsef.xx.internal.cloudapp.net/
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Co-developed-by: Souradeep Chakrabarti <schakrabarti@linux.microsoft.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Currently the teardown/setup flow for driver probe/remove is quite
a bit different from the reset flows in pdsc_fw_down()/pdsc_fw_up().
One key piece that's missing are the calls to pci_alloc_irq_vectors()
and pci_free_irq_vectors(). The pcie reset case is calling
pci_free_irq_vectors() on reset_prepare, but not calling the
corresponding pci_alloc_irq_vectors() on reset_done. This is causing
unexpected/unwanted interrupt behavior due to the adminq interrupt
being accidentally put into legacy interrupt mode. Also, the
pci_alloc_irq_vectors()/pci_free_irq_vectors() functions are being
called directly in probe/remove respectively.
Fix this inconsistency by making the following changes:
1. Always call pdsc_dev_init() in pdsc_setup(), which calls
pci_alloc_irq_vectors() and get rid of the now unused
pds_dev_reinit().
2. Always free/clear the pdsc->intr_info in pdsc_teardown()
since this structure will get re-alloced in pdsc_setup().
3. Move the calls of pci_free_irq_vectors() to pdsc_teardown()
since pci_alloc_irq_vectors() will always be called in
pdsc_setup()->pdsc_dev_init() for both the probe/remove and
reset flows.
4. Make sure to only create the debugfs "identity" entry when it
doesn't already exist, which it will in the reset case because
it's already been created in the initial call to pdsc_dev_init().
Fixes: ffa55858330f ("pds_core: implement pci reset handlers")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240129234035.69802-7-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
During reset the BARs might be accessed when they are
unmapped. This can cause unexpected issues, so fix it by
clearing the cached BAR values so they are not accessed
until they are re-mapped.
Also, make sure any places that can access the BARs
when they are NULL are prevented.
Fixes: 49ce92fbee0b ("pds_core: add FW update feature to devlink")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://lore.kernel.org/r/20240129234035.69802-6-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
There are multiple paths that can result in using the pdsc's
adminq.
[1] pdsc_adminq_isr and the resulting work from queue_work(),
i.e. pdsc_work_thread()->pdsc_process_adminq()
[2] pdsc_adminq_post()
When the device goes through reset via PCIe reset and/or
a fw_down/fw_up cycle due to bad PCIe state or bad device
state the adminq is destroyed and recreated.
A NULL pointer dereference can happen if [1] or [2] happens
after the adminq is already destroyed.
In order to fix this, add some further state checks and
implement reference counting for adminq uses. Reference
counting was used because multiple threads can attempt to
access the adminq at the same time via [1] or [2]. Additionally,
multiple clients (i.e. pds-vfio-pci) can be using [2]
at the same time.
The adminq_refcnt is initialized to 1 when the adminq has been
allocated and is ready to use. Users/clients of the adminq
(i.e. [1] and [2]) will increment the refcnt when they are using
the adminq. When the driver goes into a fw_down cycle it will
set the PDSC_S_FW_DEAD bit and then wait for the adminq_refcnt
to hit 1. Setting the PDSC_S_FW_DEAD before waiting will prevent
any further adminq_refcnt increments. Waiting for the
adminq_refcnt to hit 1 allows for any current users of the adminq
to finish before the driver frees the adminq. Once the
adminq_refcnt hits 1 the driver clears the refcnt to signify that
the adminq is deleted and cannot be used. On the fw_up cycle the
driver will once again initialize the adminq_refcnt to 1 allowing
the adminq to be used again.
Fixes: 01ba61b55b20 ("pds_core: Add adminq processing and commands")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://lore.kernel.org/r/20240129234035.69802-5-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The initial design for the adminq interrupt was done based
on client drivers having their own adminq and adminq
interrupt. So, each client driver's adminq isr would use
their specific adminqcq for the private data struct. For the
time being the design has changed to only use a single
adminq for all clients. So, instead use the struct pdsc for
the private data to simplify things a bit.
This also has the benefit of not dereferencing the adminqcq
to access the pdsc struct when the PDSC_S_STOPPING_DRIVER bit
is set and the adminqcq has actually been cleared/freed.
Fixes: 01ba61b55b20 ("pds_core: Add adminq processing and commands")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://lore.kernel.org/r/20240129234035.69802-4-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
There is a small window where pdsc_work_thread()
calls pdsc_process_adminq() and pdsc_process_adminq()
passes the PDSC_S_STOPPING_DRIVER check and starts
to process adminq/notifyq work and then the driver
starts a fw_down cycle. This could cause some
undefined behavior if the notifyqcq/adminqcq are
free'd while pdsc_process_adminq() is running. Use
cancel_work_sync() on the adminqcq's work struct
to make sure any pending work items are cancelled
and any in progress work items are completed.
Also, make sure to not call cancel_work_sync() if
the work item has not be initialized. Without this,
traces will happen in cases where a reset fails and
teardown is called again or if reset fails and the
driver is removed.
Fixes: 01ba61b55b20 ("pds_core: Add adminq processing and commands")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://lore.kernel.org/r/20240129234035.69802-3-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The PCIe reset handlers can run at the same time as the
health thread. This can cause the health thread to
stomp on the PCIe reset. Fix this by preventing the
health thread from running while a PCIe reset is happening.
As part of this use timer_shutdown_sync() during reset and
remove to make sure the timer doesn't ever get rearmed.
Fixes: ffa55858330f ("pds_core: implement pci reset handlers")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://lore.kernel.org/r/20240129234035.69802-2-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
lan743x_ptp_request_tx_timestamp uses spin_lock_bh, but it is
only called from lan743x_tx_xmit_frame where all IRQs are
already disabled.
This fixes the "IRQs not enabled as expected" warning.
Signed-off-by: Lucas Tanure <tanure@linux.com>
Link: https://lore.kernel.org/r/20240128101849.107298-1-tanure@linux.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The company I2SE has been acquired a long time ago. Switch to
my private mail address before the I2SE account is deactivated.
Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
According to MODULE_LICENSE the driver is under a dual license.
So replace the BSD license text with the proper SPDX tag.
Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
All known SPI registers of the QCA700x are 16 bit long. So adjust
the formater width accordingly.
Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Most of the users doesn't know the expected signature of the QCA700x.
So provide it within the error message. Btw use lowercase for hex as
in the rest of the driver.
Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
