Add TPH support to the Broadcom BNXT device driver. This allows the
driver to utilize TPH functions for retrieving and configuring Steering
Tags when changing interrupt affinity. With compatible NIC firmware,
network traffic will be tagged correctly with Steering Tags, resulting
in significant memory bandwidth savings and other advantages as
demonstrated by real network benchmarks on TPH-capable platforms.
Co-developed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Co-developed-by: Wei Huang <wei.huang2@amd.com>
Signed-off-by: Wei Huang <wei.huang2@amd.com>
Signed-off-by: Manoj Panicker <manoj.panicker2@amd.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-12-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In order to use queue_stop/queue_start to support the new Steering
Tags, we need to free the TX ring and TX completion ring if it is a
combined channel with TX/RX sharing the same NAPI. Otherwise
TX completions will not have the updated Steering Tag. If TPH is
not enabled, we just stop the TX ring without freeing the TX/TX cmpl
rings. With that we can now add napi_disable() and napi_enable()
during queue_stop()/ queue_start(). This will guarantee that NAPI
will stop processing the completion entries in case there are
additional pending entries in the completion rings after queue_stop().
There could be some NQEs sitting unprocessed while NAPI is disabled
thereby leaving the NQ unarmed. Explicitly re-arm the NQ after
napi_enable() in queue start so that NAPI will resume properly.
Error handling in bnxt_queue_start() requires a reset. If a TX
ring cannot be allocated or initialized properly, it will cause
TX timeout. The reset will also free any partially allocated
rings. We don't expect to hit this error path because re-allocating
previously reserved and allocated rings with the same parameters
should never fail.
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-11-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new bnxt_hwrm_tx_ring_free() function to handle freeing a HW
transmit ring. The new function will also be used in the next patch
to free the TX ring in queue_stop.
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-10-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In order to program the correct Steering Tag during an IRQ affinity
change, we need to free/re-allocate the RX completion ring during
queue_restart. If TPH is enabled, call FW to free the Rx completion
ring and clear the ring entries in queue_stop(). Re-allocate it in
queue_start() if TPH is enabled. Note that TPH mode is not enabled
in this patch and will be enabled later in the patch series.
While modifying bnxt_queue_start(), remove the unnecessary zeroing of
rxr->rx_next_cons. It gets overwritten by the clone in
bnxt_queue_start(). Remove the rx_reset counter increment since
restart is not reset. Add comment to clarify that the ring
allocations in queue_start should never fail.
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-9-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Newer firmware can use the NQ ring ID associated with each RX/RX AGG
ring to enable PCIe Steering Tags on P5_PLUS chips. When allocating
RX/RX AGG rings, pass along NQ ring ID for the firmware to use. This
information helps optimize DMA writes by directing them to the cache
closer to the CPU consuming the data, potentially improving the
processing speed. This change is backward-compatible with older
firmware, which will simply disregard the information.
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-8-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
There is some common code for setting up RX and RX AGG ring allocation
parameters for P5_PLUS chips. Refactor the logic into a new function.
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-7-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Modify bnxt_free_tx_rings() to free the skbs per TX ring.
This will be useful later in the series.
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-6-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a wrapper routine to free L2 completion rings. This will be
useful later in the series.
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-5-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new bnxt_hwrm_tx_ring_alloc() function to handle allocating
a transmit ring. This will be useful later in the series.
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-4-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new bnxt_hwrm_cp_ring_alloc_p5() function to handle allocating
one completion ring on P5_PLUS chips. This simplifies the existing code
and will be useful later in the series.
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-3-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
NPAR (Network interface card partitioning)[1] 1.2 adds a transparent
VLAN tag for all packets between the NIC and the switch. Because of
that, RX VLAN acceleration cannot be supported for any additional
host configured VLANs. The driver has to acknowledge that it can
support no RX VLAN acceleration and set the NPAR 1.2 supported flag
when registering with the FW. Otherwise, the FW call will fail and
the driver will abort on these NPAR 1.2 NICs with this error:
bnxt_en 0000:26:00.0 (unnamed net_device) (uninitialized): hwrm req_type 0x1d seq id 0xb error 0x2
[1] https://techdocs.broadcom.com/us/en/storage-and-ethernet-connectivity/ethernet-nic-controllers/bcm957xxx/adapters/introduction/features/network-partitioning-npar.html
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250213011240.1640031-2-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Lighter that normal, but the now usual collection of driver fixes and
small improvements:
- Small fixes and minor improvements to cxgb4, bnxt_re, rxe, srp, efa,
cxgb4
- Update mlx4 to use the new umem APIs, avoiding direct use of scatterlist
- Support ROCEv2 in erdma
- Remove various uncalled functions, constify bin_attribute
- Provide core infrastructure to catch netdev events and route them to
drivers, consolidating duplicated driver code
- Fix rare race condition crashes in mlx5 ODP flows
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRRRCHOFoQz/8F5bUaFwuHvBreFYQUCZ5JkNAAKCRCFwuHvBreF
YeCwAP9nrzFZMBEa0DVx7V8w3sotQCOzaaoi+4UOigeleppKSAD+K7QA5CIwNf7j
x0apvJlNsXKSYJVUI2rch/qIL8ckQwM=
=wIYu
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"Lighter that normal, but the now usual collection of driver fixes and
small improvements:
- Small fixes and minor improvements to cxgb4, bnxt_re, rxe, srp,
efa, cxgb4
- Update mlx4 to use the new umem APIs, avoiding direct use of
scatterlist
- Support ROCEv2 in erdma
- Remove various uncalled functions, constify bin_attribute
- Provide core infrastructure to catch netdev events and route them
to drivers, consolidating duplicated driver code
- Fix rare race condition crashes in mlx5 ODP flows"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (63 commits)
RDMA/mlx5: Fix implicit ODP use after free
RDMA/mlx5: Fix a race for an ODP MR which leads to CQE with error
RDMA/qib: Constify 'struct bin_attribute'
RDMA/hfi1: Constify 'struct bin_attribute'
RDMA/rxe: Fix the warning "__rxe_cleanup+0x12c/0x170 [rdma_rxe]"
RDMA/cxgb4: Notify rdma stack for IB_EVENT_QP_LAST_WQE_REACHED event
RDMA/bnxt_re: Allocate dev_attr information dynamically
RDMA/bnxt_re: Pass the context for ulp_irq_stop
RDMA/bnxt_re: Add support to handle DCB_CONFIG_CHANGE event
RDMA/bnxt_re: Query firmware defaults of CC params during probe
RDMA/bnxt_re: Add Async event handling support
bnxt_en: Add ULP call to notify async events
RDMA/mlx5: Fix indirect mkey ODP page count
MAINTAINERS: Update the bnxt_re maintainers
RDMA/hns: Clean up the legacy CONFIG_INFINIBAND_HNS
RDMA/rtrs: Add missing deinit() call
RDMA/efa: Align interrupt related fields to same type
RDMA/bnxt_re: Fix to drop reference to the mmap entry in case of error
RDMA/mlx5: Fix link status down event for MPV
RDMA/erdma: Support create_ah/destroy_ah in non-sleepable contexts
...
300-400B RPC requests are fairly common. With the current default
of 256B HDS threshold bnxt ends up splitting those, lowering PCIe
bandwidth efficiency and increasing the number of memory allocation.
Increase the HDS threshold to fit 4 buffers in a 4k page.
This works out to 640B as the threshold on a typical kernel confing.
This change increases the performance for a microbenchmark which
receives 400B RPCs and sends empty responses by 4.5%.
Admittedly this is just a single benchmark, but 256B works out to
just 6 (so 2 more) packets per head page, because shinfo size
dominates the headers.
Now that we use page pool for the header pages I was also tempted
to default rx_copybreak to 0, but in synthetic testing the copybreak
size doesn't seem to make much difference.
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250119020518.1962249-8-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Now that we can configure HDS threshold separately from the rx_copybreak
HDS threshold may be higher than rx_copybreak.
We need to make sure that we have enough space for the headers.
Fixes: 6b43673a25 ("bnxt_en: add support for hds-thresh ethtool command")
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250119020518.1962249-7-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the pending config for hds_thrs. Core will only update the "current"
one after we return success. Without this change 2 reconfigs would be
required for the setting to reach the device.
Fixes: 6b43673a25 ("bnxt_en: add support for hds-thresh ethtool command")
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250119020518.1962249-5-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Separate the HDS config from the ethtool state struct.
The HDS config contains just simple parameters, not state.
Having it as a separate struct will make it easier to clone / copy
and also long term potentially make it per-queue.
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250119020518.1962249-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The bnxt_en driver has configured the hds_threshold value automatically
when TPA is enabled based on the rx-copybreak default value.
Now the hds-thresh ethtool command is added, so it adds an
implementation of hds-thresh option.
Configuration of the hds-thresh is applied only when
the tcp-data-split is enabled. The default value of
hds-thresh is 256, which is the default value of
rx-copybreak, which used to be the hds_thresh value.
The maximum hds-thresh is 1023.
# Example:
# ethtool -G enp14s0f0np0 tcp-data-split on hds-thresh 256
# ethtool -g enp14s0f0np0
Ring parameters for enp14s0f0np0:
Pre-set maximums:
...
HDS thresh: 1023
Current hardware settings:
...
TCP data split: on
HDS thresh: 256
Tested-by: Stanislav Fomichev <sdf@fomichev.me>
Tested-by: Andy Gospodarek <gospo@broadcom.com>
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250114142852.3364986-9-ap420073@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
NICs that uses bnxt_en driver supports tcp-data-split feature by the
name of HDS(header-data-split).
But there is no implementation for the HDS to enable by ethtool.
Only getting the current HDS status is implemented and The HDS is just
automatically enabled only when either LRO, HW-GRO, or JUMBO is enabled.
The hds_threshold follows rx-copybreak value. and it was unchangeable.
This implements `ethtool -G <interface name> tcp-data-split <value>`
command option.
The value can be <on> and <auto>.
The value is <auto> and one of LRO/GRO/JUMBO is enabled, HDS is
automatically enabled and all LRO/GRO/JUMBO are disabled, HDS is
automatically disabled.
HDS feature relies on the aggregation ring.
So, if HDS is enabled, the bnxt_en driver initializes the aggregation ring.
This is the reason why BNXT_FLAG_AGG_RINGS contains HDS condition.
Acked-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Stanislav Fomichev <sdf@fomichev.me>
Tested-by: Andy Gospodarek <gospo@broadcom.com>
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250114142852.3364986-8-ap420073@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The bnxt_en driver supports rx-copybreak, but it couldn't be set by
userspace. Only the default value(256) has worked.
This patch makes the bnxt_en driver support following command.
`ethtool --set-tunable <devname> rx-copybreak <value> ` and
`ethtool --get-tunable <devname> rx-copybreak`.
By this patch, hds_threshol is set to the rx-copybreak value.
But it will be set by `ethtool -G eth0 hds-thresh N`
in the next patch.
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Tested-by: Stanislav Fomichev <sdf@fomichev.me>
Tested-by: Andy Gospodarek <gospo@broadcom.com>
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250114142852.3364986-7-ap420073@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the driver receives an async event notification from the Firmware,
we make the new ulp_async_notifier() call to inform the RDMA driver that
a firmware async event has been received. RDMA driver can then take
necessary actions based on the event type.
In the next patch, we will implement the ulp_async_notifier() callbacks
in the RDMA driver.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Link: https://patch.msgid.link/20250107024553.2926983-2-kalesh-anakkur.purayil@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Recalculate features when XDP is detached.
Before:
# ip li set dev eth0 xdp obj xdp_dummy.bpf.o sec xdp
# ip li set dev eth0 xdp off
# ethtool -k eth0 | grep gro
rx-gro-hw: off [requested on]
After:
# ip li set dev eth0 xdp obj xdp_dummy.bpf.o sec xdp
# ip li set dev eth0 xdp off
# ethtool -k eth0 | grep gro
rx-gro-hw: on
The fact that HW-GRO doesn't get re-enabled automatically is just
a minor annoyance. The real issue is that the features will randomly
come back during another reconfiguration which just happens to invoke
netdev_update_features(). The driver doesn't handle reconfiguring
two things at a time very robustly.
Starting with commit 98ba1d931f ("bnxt_en: Fix RSS logic in
__bnxt_reserve_rings()") we only reconfigure the RSS hash table
if the "effective" number of Rx rings has changed. If HW-GRO is
enabled "effective" number of rings is 2x what user sees.
So if we are in the bad state, with HW-GRO re-enablement "pending"
after XDP off, and we lower the rings by / 2 - the HW-GRO rings
doing 2x and the ethtool -L doing / 2 may cancel each other out,
and the:
if (old_rx_rings != bp->hw_resc.resv_rx_rings &&
condition in __bnxt_reserve_rings() will be false.
The RSS map won't get updated, and we'll crash with:
BUG: kernel NULL pointer dereference, address: 0000000000000168
RIP: 0010:__bnxt_hwrm_vnic_set_rss+0x13a/0x1a0
bnxt_hwrm_vnic_rss_cfg_p5+0x47/0x180
__bnxt_setup_vnic_p5+0x58/0x110
bnxt_init_nic+0xb72/0xf50
__bnxt_open_nic+0x40d/0xab0
bnxt_open_nic+0x2b/0x60
ethtool_set_channels+0x18c/0x1d0
As we try to access a freed ring.
The issue is present since XDP support was added, really, but
prior to commit 98ba1d931f ("bnxt_en: Fix RSS logic in
__bnxt_reserve_rings()") it wasn't causing major issues.
Fixes: 1054aee823 ("bnxt_en: Use NETIF_F_GRO_HW.")
Fixes: 98ba1d931f ("bnxt_en: Fix RSS logic in __bnxt_reserve_rings()")
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Link: https://patch.msgid.link/20250109043057.2888953-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
DIM work will call the firmware to adjust the coalescing parameters on
the RX rings. We should cancel DIM work before we call the firmware
to free the RX rings. Otherwise, FW will reject the call from DIM
work if the RX ring has been freed. This will generate an error
message like this:
bnxt_en 0000:21:00.1 ens2f1np1: hwrm req_type 0x53 seq id 0x6fca error 0x2
and cause unnecessary concern for the user. It is also possible to
modify the coalescing parameters of the wrong ring if the ring has
been re-allocated.
To prevent this, cancel DIM work right before freeing the RX rings.
We also have to add a check in NAPI poll to not schedule DIM if the
RX rings are shutting down. Check that the VNIC is active before we
schedule DIM. The VNIC is always disabled before we free the RX rings.
Fixes: 0bc0b97fca ("bnxt_en: cleanup DIM work on device shutdown")
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20250104043849.3482067-3-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Call the new HWRM_PORT_MAC_QCAPS to check if mac loopback is
supported. Skip the MAC loopback ethtool self test if it is
not supported.
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Link: https://patch.msgid.link/20241217182620.2454075-5-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Block all ethtool module operations on an untrusted VF. The firmware
won't allow it and will return error.
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241217182620.2454075-3-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If FW supports setting resource limits for RoCE, then just use the
FW limits instead of using some fixed values in the driver. These
limits will be used to allocate context memory for QP, SRQ, AH, and
MR resources for RoCE.
Reviewed-by: Damodharam Ammepalli <damodharam.ammepalli@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241217182620.2454075-2-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The existing code is using RSS profile to determine IPV4/IPV6 GSO type
on all chips older than 5760X. This won't work on 5750X chips that may
be using modified RSS profiles. This commit from 2018 has updated the
driver to not use RSS profile for HW GRO packets on newer chips:
50f011b63d ("bnxt_en: Update RSS setup and GRO-HW logic according to the latest spec.")
However, a recent commit to add support for the newest 5760X chip broke
the logic. If the GRO packet needs to be re-segmented by the stack, the
wrong GSO type will cause the packet to be dropped.
Fix it to only use RSS profile to determine GSO type on the oldest
5730X/5740X chips which cannot use the new method and is safe to use the
RSS profiles.
Also fix the L3/L4 hash type for RX packets by not using the RSS
profile for the same reason. Use the ITYPE field in the RX completion
to determine L3/L4 hash types correctly.
Fixes: a7445d6980 ("bnxt_en: Add support for new RX and TPA_START completion types for P7")
Reviewed-by: Colin Winegarden <colin.winegarden@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241204215918.1692597-2-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 7ed816be35 ("eth: bnxt: use page pool for head frags") added a
page pool for header frags, which may be distinct from the existing pool
for the aggregation ring. Prior to this change, frags used in the TPA
ring rx_tpa were allocated from system memory e.g. napi_alloc_frag()
meaning their lifetimes were not associated with a page pool. They can
be returned at any time and so the queue API did not alloc or free
rx_tpa.
But now frags come from a separate head_pool which may be different to
page_pool. Without allocating and freeing rx_tpa, frags allocated from
the old head_pool may be returned to a different new head_pool which
causes a mismatch between the pp hold/release count.
Fix this problem by properly freeing and allocating rx_tpa in the queue
API implementation.
Signed-off-by: David Wei <dw@davidwei.uk>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241204041022.56512-4-dw@davidwei.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Refactor bnxt_rx_ring_info->tpa_info operations into helpers that work
on a single tpa_info in prep for queue API using them.
There are 2 pairs of operations:
* bnxt_alloc_one_tpa_info()
* bnxt_free_one_tpa_info()
These alloc/free the tpa_info array itself.
* bnxt_alloc_one_tpa_info_data()
* bnxt_free_one_tpa_info_data()
These alloc/free the frags stored in tpa_info array.
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: David Wei <dw@davidwei.uk>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241204041022.56512-2-dw@davidwei.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If we go through the PCI shutdown or suspend path, we shutdown the
NIC but PTP remains registered. If the kernel continues to run for
a little bit, the periodic PTP .do_aux_work() function may be called
and it will read the PHC from the BAR register. Since the device
has already been disabled, it will cause a PCIe completion timeout.
Fix it by calling bnxt_ptp_clear() in the PCI shutdown/suspend
handlers. bnxt_ptp_clear() will unregister from PTP and
.do_aux_work() will be canceled.
In bnxt_resume(), we need to re-initialize PTP.
Fixes: a521c8a01d ("bnxt_en: Move bnxt_ptp_init() from bnxt_open() back to bnxt_init_one()")
Cc: Richard Cochran <richardcochran@gmail.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Instead of passing the 2nd parameter phc_cfg to bnxt_ptp_init().
Store it in bp->ptp_cfg so that the caller doesn't need to know what
the value should be.
In the next patch, we'll need to call bnxt_ptp_init() in bnxt_resume()
and this will make it easier.
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The MTU setting at the time an XDP multi-buffer is attached
determines whether the aggregation ring will be used and the
rx_skb_func handler. This is done in bnxt_set_rx_skb_mode().
If the MTU is later changed, the aggregation ring setting may need
to be changed and it may become out-of-sync with the settings
initially done in bnxt_set_rx_skb_mode(). This may result in
random memory corruption and crashes as the HW may DMA data larger
than the allocated buffer size, such as:
BUG: kernel NULL pointer dereference, address: 00000000000003c0
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 17 PID: 0 Comm: swapper/17 Kdump: loaded Tainted: G S OE 6.1.0-226bf9805506 #1
Hardware name: Wiwynn Delta Lake PVT BZA.02601.0150/Delta Lake-Class1, BIOS F0E_3A12 08/26/2021
RIP: 0010:bnxt_rx_pkt+0xe97/0x1ae0 [bnxt_en]
Code: 8b 95 70 ff ff ff 4c 8b 9d 48 ff ff ff 66 41 89 87 b4 00 00 00 e9 0b f7 ff ff 0f b7 43 0a 49 8b 95 a8 04 00 00 25 ff 0f 00 00 <0f> b7 14 42 48 c1 e2 06 49 03 95 a0 04 00 00 0f b6 42 33f
RSP: 0018:ffffa19f40cc0d18 EFLAGS: 00010202
RAX: 00000000000001e0 RBX: ffff8e2c805c6100 RCX: 00000000000007ff
RDX: 0000000000000000 RSI: ffff8e2c271ab990 RDI: ffff8e2c84f12380
RBP: ffffa19f40cc0e48 R08: 000000000001000d R09: 974ea2fcddfa4cbf
R10: 0000000000000000 R11: ffffa19f40cc0ff8 R12: ffff8e2c94b58980
R13: ffff8e2c952d6600 R14: 0000000000000016 R15: ffff8e2c271ab990
FS: 0000000000000000(0000) GS:ffff8e3b3f840000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000000003c0 CR3: 0000000e8580a004 CR4: 00000000007706e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
<IRQ>
__bnxt_poll_work+0x1c2/0x3e0 [bnxt_en]
To address the issue, we now call bnxt_set_rx_skb_mode() within
bnxt_change_mtu() to properly set the AGG rings configuration and
update rx_skb_func based on the new MTU value.
Additionally, BNXT_FLAG_NO_AGG_RINGS is cleared at the beginning of
bnxt_set_rx_skb_mode() to make sure it gets set or cleared based on
the current MTU.
Fixes: 08450ea98a ("bnxt_en: Fix max_mtu setting for multi-buf XDP")
Co-developed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Shravya KN <shravya.k-n@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
HWRM_RING_FREE followed by a HWRM_RING_ALLOC is not guaranteed to
have the same FW ring ID as before. So we must reinitialize the
RSS table with the correct ring IDs. Otherwise, traffic may not
resume properly if the restarted ring ID is stale. Since this
feature is only supported on P5_PLUS chips, we call
bnxt_vnic_set_rss_p5() to update the HW RSS table.
Fixes: 2d694c27d3 ("bnxt_en: implement netdev_queue_mgmt_ops")
Cc: David Wei <dw@davidwei.uk>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
After successful PCIe AER recovery, FW will reset all resource
reservations. If it is IF_UP, the driver will call bnxt_open() and
all resources will be reserved again. It it is IF_DOWN, we should
call bnxt_reserve_rings() so that we can reserve resources including
RoCE resources to allow RoCE to resume after AER. Without this
patch, RoCE fails to resume in this IF_DOWN scenario.
Later, if it becomes IF_UP, bnxt_open() will see that resources have
been reserved and will not reserve again.
Fixes: fb1e6e562b ("bnxt_en: Fix AER recovery.")
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Seveal fixes scattered across the drivers and a few new features:
- Minor updates and bug fixes to hfi1, efa, iopob, bnxt, hns
- Force disassociate the userspace FD when hns does an async reset
- bnxt new features for optimized modify QP to skip certain stayes, CQ
coalescing, better debug dumping
- mlx5 new data placement ordering feature
- Faster destruction of mlx5 devx HW objects
- Improvements to RDMA CM mad handling
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRRRCHOFoQz/8F5bUaFwuHvBreFYQUCZz4ENwAKCRCFwuHvBreF
YQYQAP9R54r5J1Iylg+zqhCc+e/9oveuuZbfLvy/EJiEpmdprQEAgPs1RrB0z7U6
1xrVStUKNPhGd5XeVVZGkIV0zYv6Tw4=
=V5xI
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"Seveal fixes scattered across the drivers and a few new features:
- Minor updates and bug fixes to hfi1, efa, iopob, bnxt, hns
- Force disassociate the userspace FD when hns does an async reset
- bnxt new features for optimized modify QP to skip certain stayes,
CQ coalescing, better debug dumping
- mlx5 new data placement ordering feature
- Faster destruction of mlx5 devx HW objects
- Improvements to RDMA CM mad handling"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (51 commits)
RDMA/bnxt_re: Correct the sequence of device suspend
RDMA/bnxt_re: Use the default mode of congestion control
RDMA/bnxt_re: Support different traffic class
IB/cm: Rework sending DREQ when destroying a cm_id
IB/cm: Do not hold reference on cm_id unless needed
IB/cm: Explicitly mark if a response MAD is a retransmission
RDMA/mlx5: Move events notifier registration to be after device registration
RDMA/bnxt_re: Cache MSIx info to a local structure
RDMA/bnxt_re: Refurbish CQ to NQ hash calculation
RDMA/bnxt_re: Refactor NQ allocation
RDMA/bnxt_re: Fail probe early when not enough MSI-x vectors are reserved
RDMA/hns: Fix different dgids mapping to the same dip_idx
RDMA/bnxt_re: Add set_func_resources support for P5/P7 adapters
RDMA/bnxt_re: Enhance RoCE SRIOV resource configuration design
bnxt_en: Add support for RoCE sriov configuration
RDMA/hns: Fix NULL pointer derefernce in hns_roce_map_mr_sg()
RDMA/hns: Fix out-of-order issue of requester when setting FENCE
RDMA/nldev: Add IB device and net device rename events
RDMA/mlx5: Add implementation for ufile_hw_cleanup device operation
RDMA/core: Move ib_uverbs_file struct to uverbs_types.h
...
Host context memory is used by the newer chips to store context
information for various L2 and RoCE states and FW logs. This
information will be useful for debugging. This patch adds the
functions to copy all pages of a context memory type to a contiguous
buffer. The next patches will include the context memory dump
during ethtool -w coredump.
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Co-developed-by: Shruti Parab <shruti.parab@broadcom.com>
Signed-off-by: Shruti Parab <shruti.parab@broadcom.com>
Signed-off-by: Sreekanth Reddy <sreekanth.reddy@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-9-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If FW supports appending new FW logs to an offset in the context
memory after FW reset, then do not free this type of context memory
during reset. The driver will provide the initial offset to the FW
when configuring this type of context memory. This way, we don't lose
the older FW logs after reset.
Signed-off-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-8-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The FW trace memory pages will be added to the ethtool -w coredump
in later patches. In addition to the raw data, the driver has to
add a header to provide the head and tail information on each FW
trace log segment when creating the coredump. The FW sends an async
message to the driver after DMAing a chunk of logs to the context
memory to indicate the last offset containing the tail of the logs.
The driver needs to keep track of that.
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Shruti Parab <shruti.parab@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-7-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allocate the new FW trace log backing store context memory types
if they are supported by the FW. FW debug logs are DMA'ed to the host
backing store memory when the on-chip buffers are full. If host
memory cannot be allocated for these memory types, the driver
will not abort.
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Shruti Parab <shruti.parab@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-6-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If 'force' is false, it will keep the memory pages and all data
structures for the context memory type if the memory is valid.
This patch always passes true for the 'force' parameter so there is
no change in behavior. Later patches will adjust the 'force' parameter
for the FW log context memory types so that the logs will not be reset
after FW reset.
Signed-off-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-5-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new function bnxt_free_one_ctx_mem() to free one context
memory type. bnxt_free_ctx_mem() now calls the new function in
the loop to free each context memory type. There is no change in
behavior. Later patches will further make use of the new function.
Signed-off-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-4-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new bit to struct bnxt_ctx_mem_type to indicate that host
memory has been successfully allocated for this context memory type.
In the next patches, we'll be adding some additional context memory
types for FW debugging/logging. If memory cannot be allocated for
any of these new types, we will not abort and the cleared mem_valid
bit will indicate to skip configuring the memory type.
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Shruti Parab <shruti.parab@broadcom.com>
Signed-of-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20241115151438.550106-3-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Testing small size RPCs (300B-400B) on a large AMD system suggests
that page pool recycling is very useful even for just the head frags.
With this patch (and copy break disabled) I see a 30% performance
improvement (82Gbps -> 106Gbps).
Convert bnxt from normal page frags to page pool frags for head buffers.
On systems with small page size we can use the same pool as for TPA
pages. On systems with large pages the frag allocation logic of the
page pool is already used to split a large page into TPA chunks.
TPA chunks are much larger than heads (8k or 64k, AFAICT vs 1kB)
and we always allocate the same sized chunks. Mixing allocation
of TPA and head pages would lead to sub-optimal memory use.
Plus Taehee's work on zero-copy / devmem will need to differentiate
between TPA and non-TPA page pool, anyway. Conditionally allocate
a new page pool for heads.
Link: https://patch.msgid.link/20241109035119.3391864-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
During driver load, PF RDMA driver provisions resources
to the RDMA VFs. This logic takes into consideration of
the total number of VFs supported on the PF while
allocating resources. Firmware now advertises a capability
where NIC driver can allocate resources for RDMA VFs when
the user actually creates a VF. So this resource
distribution can be based on the number of active VFs.
This patch adds the support to check for the firmware
capability and follow the new RDMA VF resource allocation
strategy. The current logic in the RDMA driver will be
removed for the newer Firmware versions in a subsequent
patch in this series.
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://patch.msgid.link/1730882676-24434-2-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
irq_set_affinity_hint() is deprecated, Use irq_update_affinity_hint()
instead. This removes the side-effect of actually applying the affinity.
The driver does not really need to worry about spreading its IRQs across
CPUs. The core code already takes care of that. when the driver applies the
affinities by itself, it breaks the users' expectations:
1. The user configures irqbalance with IRQBALANCE_BANNED_CPULIST in
order to prevent IRQs from being moved to certain CPUs that run a
real-time workload.
2. bnxt_en device reopening will resets the affinity
in bnxt_open().
3. bnxt_en has no idea about irqbalance's config, so it may move an IRQ to
a banned CPU. The real-time workload suffers unacceptable latency.
Signed-off-by: Mohammad Heib <mheib@redhat.com>
Reviewed-by: Andy Gospodarek <gospo@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Link: https://patch.msgid.link/20241106180811.385175-1-mheib@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We can see high contention on ptp_lock while doing RX timestamping
on high packet rates over several queues. Spinlock is not effecient
to protect timecounter for RX timestamps when reads are the most
usual operations and writes are only occasional. It's better to use
seqlock in such cases.
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Vadim Fedorenko <vadfed@meta.com>
Link: https://patch.msgid.link/20241103215108.557531-2-vadfed@meta.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
