| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
Mellanox, mlx5 fixes 2019-04-09
This series provides some fixes to mlx5 driver.
I've cc'ed some of the checksum fixes to Eric Dumazet and i would like to get
his feedback before you pull.
For -stable v4.19
('net/mlx5: FPGA, tls, idr remove on flow delete')
('net/mlx5: FPGA, tls, hold rcu read lock a bit longer')
For -stable v4.20
('net/mlx5e: Rx, Check ip headers sanity')
('Revert "net/mlx5e: Enable reporting checksum unnecessary also for L3 packets"')
('net/mlx5e: Rx, Fixup skb checksum for packets with tail padding')
For -stable v5.0
('net/mlx5e: Switch to Toeplitz RSS hash by default')
('net/mlx5e: Protect against non-uplink representor for encap')
('net/mlx5e: XDP, Avoid checksum complete when XDP prog is loaded')
====================
Signed-off-by: David S. Miller <[email protected]>
|
|
Separate the overflow handling from the hardware interrupt status analysis.
The interrupt status is a single register and is common for all PFs. The
first PF reading the register is not necessarily the one who overflowed.
All PFs must check their overflow status on every attention.
In this change we clear the sticky indication in the attention handler to
allow doorbells to be processed again as soon as possible, but running
the doorbell recovery is scheduled for the periodic handler to reduce the
time spent in the attention handler.
Checking the need for DORQ flush was changed to "db_bar_no_edpm" because
qed_edpm_enabled()'s result could change dynamically and might have
prevented a needed flush.
Signed-off-by: Denis Bolotin <[email protected]>
Signed-off-by: Michal Kalderon <[email protected]>
Signed-off-by: Ariel Elior <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When the DORQ (doorbell block) is overflowed, all PFs get attentions at the
same time. If one PF finished handling the attention before another PF even
started, the second PF might miss the DORQ's attention bit and not handle
the attention at all.
If the DORQ attention is missed and the issue is not resolved, another
attention will not be sent, therefore each attention is treated as a
potential DORQ attention.
As a result, the attention callback is called more frequently so the debug
print was moved to reduce its quantity.
The number of periodic doorbell recovery handler schedules was reduced
because it was the previous way to mitigating the missed attention issue.
Signed-off-by: Denis Bolotin <[email protected]>
Signed-off-by: Michal Kalderon <[email protected]>
Signed-off-by: Ariel Elior <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Fix the condition which verifies that doorbell address is inside the
doorbell bar by checking that the end of the address is within range
as well.
Signed-off-by: Denis Bolotin <[email protected]>
Signed-off-by: Michal Kalderon <[email protected]>
Signed-off-by: Ariel Elior <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
DB_REC_DRY_RUN (running doorbell recovery without sending doorbells) is
never used. DB_REC_ONCE (send a single doorbell from the doorbell recovery)
is not needed anymore because by running the periodic handler we make sure
we check the overflow status later instead.
This patch is needed because in the next patches, the only doorbell
recovery type being used is DB_REC_REAL_DEAL, and the fixes are much
cleaner without this enum.
Signed-off-by: Denis Bolotin <[email protected]>
Signed-off-by: Michal Kalderon <[email protected]>
Signed-off-by: Ariel Elior <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This check isn't really needed and we can simplify the code and save
some CPU cycles by removing it. Only in case of an error none of these
bits are set, and calling the NAPI callback doesn't hurt in this case.
Signed-off-by: Heiner Kallweit <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Using a function pointer array makes this easier to read and better
maintainable.
Signed-off-by: Heiner Kallweit <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Using a function pointer array makes this easier to read and better
maintainable. AFAIK function pointer arrays cause some performance
drawback due to Spectre mitigation, but we're not in a hot path here.
Signed-off-by: Heiner Kallweit <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This patch removes some redundant BH disable when initializing
and uninitializing command queue.
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
If there is pending skb in RX flow when close the port, and the
pending buffer is not cleaned, the new packet will be added to
the pending skb when the port opens again, and the first new
packet has error data.
This patch cleans the pending skb when clean RX ring.
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
For some cases, PHY may not be connected to MDIO bus, then
the driver will initialize fail since MDIO bus initialization
fails.
This patch fixes it by skipping the MDIO bus initialization
when PHY is inexistent.
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
According to hardware description, reset level that should be
triggered are not consistent in a module. For example, in SSU
common errors, the first two bits has no need to do reset,
but the other bits need global reset.
This patch sets separate reset level for all RAS and MSI-X
interrupts by adding a reset_lvel field in struct hclge_hw_error,
and fixes some incorrect reset level.
Signed-off-by: Weihang Li <[email protected]>
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Currently hardware may have not enough buffer to receive packet
when it has used more than two MPS(maximum packet size) of
buffer, but there are still a lot of shared buffer left unused
when TC num is small.
This patch divides shared buffer to be used between TC when
the port supports DCB, and adjusts the waterline and threshold
according to user manual for the port that does not support
DCB.
This patch also change hclge_get_tc_num's return type to u32
to avoid signed-unsigned mix with divide.
Signed-off-by: Yunsheng Lin <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Currently RX shared buffer' threshold size for speific TC is
set to smaller value when the TC's PFC is not enabled, which may
cause performance problem because hardware may not have enough
hardware buffer when PFC is not enabled.
This patch sets the same threshold size for all TC no matter if
the specific TC's PFC is enabled.
Signed-off-by: Yunsheng Lin <[email protected]>
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When a GRO packet is received by driver, the cwr field in the
struct tcphdr needs to be checked to decide whether to set the
SKB_GSO_TCP_ECN for skb_shinfo(skb)->gso_type.
So this patch adds hns3_gro_complete to do that, and adds the
hns3_handle_bdinfo to handle the hns3_gro_complete and
hns3_rx_checksum.
Signed-off-by: Yunsheng Lin <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Change the parameters of hns3_rx_checksum to be more specific to
what is used internally, rather than passing in a pointer to the
whole hns3_desc. Reduces duplicate code and bring this function
inline with the approach used in hns3_set_gro_param.
Signed-off-by: Yunsheng Lin <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In original codes, ndo_set_vf_vlan() in hns3 driver was implemented
wrong. It adds or removes VLAN into VLAN filter for VF, but VF is
unaware of it.
This patch fixes it. When VF loads up, it firstly queries the port
based VLAN state from PF. When user change port based VLAN state
from PF, PF firstly checks whether the VF is alive. If the VF is
alive, then PF notifies the VF the modification; otherwise PF
configure the port based VLAN state directly.
Fixes: 46a3df9f9718 ("net: hns3: Add HNS3 Acceleration Engine & Compatibility Layer Support")
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In original codes, ndo_set_vf_vlan() in hns3 driver was implemented
wrong. It adds or removes VLAN into VLAN filter for VF, but VF is
unaware of it.
Indeed, ndo_set_vf_vlan() is expected to enable or disable port based
VLAN (hardware inserts a specified VLAN tag to all TX packets for a
specified VF) . When enable port based VLAN, we use port based VLAN id
as VLAN filter entry. When disable port based VLAN, we use VLAN id of
VLAN device.
This patch fixes it for PF, enable/disable port based VLAN when calls
ndo_set_vf_vlan().
Fixes: 46a3df9f9718 ("net: hns3: Add HNS3 Acceleration Engine & Compatibility Layer Support")
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Currently, in TX direction, driver implements the TX VLAN offload
by checking the VLAN header in skb, and filling it into TX descriptor.
Usually it works well, but if enable inserting VLAN header based on
port, it may conflict when out_tag field of TX descriptor is already
used, and cause RAS error.
In RX direction, hardware supports stripping max two VLAN headers.
For vlan_tci in skb can only store one VLAN tag, when RX VLAN offload
enabled, driver tells hardware to strip one VLAN header from RX
packet; when RX VLAN offload disabled, driver tells hardware not to
strip VLAN header from RX packet. Now if port based insert VLAN
enabled, all RX packets will have the port based VLAN header. This
header is useless for stack, driver needs to ask hardware to strip
it. Unfortunately, hardware can't drop this VLAN header, and always
fill it into RX descriptor, so driver has to identify and drop it.
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Our hardware supports inserting a specified VLAN header for each
function when sending packets. User can enable it with command
"ip link set <devname> vf <vfid> vlan <vlan id>".
For this VLAN header is inserted by hardware, not from stack,
hardware also needs to strip it from received packets before
sending to stack. In this case, driver needs to tell
hardware which VLAN to insert or strip.
The current VLAN initialization doesn't allow inserting
VLAN header by hardware, this patch modifies it, in order be
compatible with VLAN inserted base on port.
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
BPF's control message handler seems like a good base to built
on for request-reply control messages. Split it out to allow
for reuse.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Dirk van der Merwe <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
During probe we clear vNIC configuration in case the device
wasn't closed cleanly by previous driver. Move that code
before netdev init, so netdev init can already try to apply
its config parameters.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Dirk van der Merwe <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Soon we will try to write to the vNIC mailbox without RTNL held.
Add a new mutex to protect access to specific parts of the PCI
control BAR.
Move the mailbox size checking to the mailbox lock() helper, where
it can be more effective (happen prior to potential overwrite of
other data).
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Dirk van der Merwe <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
If the reconfig was a quick update, we could have results available from
firmware within 200us.
Signed-off-by: Dirk van der Merwe <[email protected]>
Signed-off-by: Jakub Kicinski <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The err2 error return path calls qede_ptp_disable that cleans up
on an error and frees ptp. After this, the free'd ptp is dereferenced
when ptp->clock is set to NULL and the code falls-through to error
path err1 that frees ptp again.
Fix this by calling qede_ptp_disable and exiting via an error
return path that does not set ptp->clock or kfree ptp.
Addresses-Coverity: ("Write to pointer after free")
Fixes: 035744975aec ("qede: Add support for PTP resource locking.")
Signed-off-by: Colin Ian King <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Currently if a pci dma mapping failure is detected a free'd
memblock address is returned rather than a NULL (that indicates
an error). Fix this by ensuring NULL is returned on this error case.
Addresses-Coverity: ("Use after free")
Fixes: 528f727279ae ("vxge: code cleanup and reorganization")
Signed-off-by: Colin Ian King <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Some SOC like i.MX6SX clock have some limits:
- ahb clock should be disabled before ipg.
- ahb and ipg clocks are required for MAC MII bus.
So, move the ahb clock to runtime management together with
ipg clock.
Signed-off-by: Fugang Duan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The thunderx driver forbids to load an eBPF program if the MTU is too high,
but this can be circumvented by loading the eBPF, then raising the MTU.
Fix this by limiting the MTU if an eBPF program is already loaded.
Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support")
Signed-off-by: Matteo Croce <[email protected]>
Acked-by: Jesper Dangaard Brouer <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The thunderx driver splits frames bigger than 1530 bytes to multiple
pages, making impossible to run an eBPF program on it.
This leads to a maximum MTU of 1508 if QinQ is in use.
The thunderx driver forbids to load an eBPF program if the MTU is higher
than 1500 bytes. Raise the limit to 1508 so it is possible to use L2
protocols which need some more headroom.
Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support")
Signed-off-by: Matteo Croce <[email protected]>
Acked-by: Jesper Dangaard Brouer <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
While determining offload capabilities of backing hardware during
a device reset, the driver is clobbering current feature settings.
Update hw_features on reset instead of features unless a feature
is enabled that is no longer supported on the current backing device.
Also enable features that were not supported prior to the reset but
were previously enabled or requested by the user.
This can occur if the reset is the result of a carrier change, such
as a device failover or partition migration.
Signed-off-by: Thomas Falcon <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Enable Generic Receive Offload in the ibmvnic driver.
Signed-off-by: Thomas Falcon <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In Spectrum-1, when a multicast packet is admitted to the shared buffer
it increases the quotas of all the ports and {port, TC} to which it is
forwarded to.
The above means that multicast packets are accounted multiple times in
the shared buffer and can therefore cause the associated shared buffer
pool to fill up very quickly.
To work around this issue, commit e83c045e53d7 ("mlxsw:
spectrum_buffers: Configure MC pool") added a dedicated multicast pool
in which multicast packets are accounted.
The issue is not present in Spectrum-2, but in order to be backward
compatible with Spectrum-1, its default behavior is to allow a multicast
packet to increase multiple egress quotas instead of one.
Until the new (non-backward compatible) mode is supported, configure a
dedicated multicast pool as in Spectrum-1.
Fixes: fe099bf682ab ("mlxsw: spectrum_buffers: Add Spectrum-2 shared buffer configuration")
Signed-off-by: Ido Schimmel <[email protected]>
Reviewed-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Commit 74bc99397438 ("mlxsw: spectrum_router: Veto unsupported RIF MAC
addresses") enabled the driver to veto router interface (RIF) MAC
addresses that it cannot support.
This check should only be performed for interfaces for which the driver
actually configures a RIF. A VRF upper is not one of them, so ignore it.
Without this patch it is not possible to set an IP address on the VRF
device and use it as a loopback.
Fixes: 74bc99397438 ("mlxsw: spectrum_router: Veto unsupported RIF MAC addresses")
Signed-off-by: Ido Schimmel <[email protected]>
Reported-by: Alexander Petrovskiy <[email protected]>
Tested-by: Alexander Petrovskiy <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The workqueue is used to periodically update the networking stack about
activity / statistics of various objects such as neighbours and TC
actions.
It should not be called as part of memory reclaim path, so remove the
WQ_MEM_RECLAIM flag.
Fixes: 3d5479e92087 ("mlxsw: core: Remove deprecated create_workqueue")
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The ordered workqueue is used to offload various objects such as routes
and neighbours in the order they are notified.
It should not be called as part of memory reclaim path, so remove the
WQ_MEM_RECLAIM flag. This can also result in a warning [1], if a worker
tries to flush a non-WQ_MEM_RECLAIM workqueue.
[1]
[97703.542861] workqueue: WQ_MEM_RECLAIM mlxsw_core_ordered:mlxsw_sp_router_fib6_event_work [mlxsw_spectrum] is flushing !WQ_MEM_RECLAIM events:rht_deferred_worker
[97703.542884] WARNING: CPU: 1 PID: 32492 at kernel/workqueue.c:2605 check_flush_dependency+0xb5/0x130
...
[97703.542988] Hardware name: Mellanox Technologies Ltd. MSN3700C/VMOD0008, BIOS 5.11 10/10/2018
[97703.543049] Workqueue: mlxsw_core_ordered mlxsw_sp_router_fib6_event_work [mlxsw_spectrum]
[97703.543061] RIP: 0010:check_flush_dependency+0xb5/0x130
...
[97703.543071] RSP: 0018:ffffb3f08137bc00 EFLAGS: 00010086
[97703.543076] RAX: 0000000000000000 RBX: ffff96e07740ae00 RCX: 0000000000000000
[97703.543080] RDX: 0000000000000094 RSI: ffffffff82dc1934 RDI: 0000000000000046
[97703.543084] RBP: ffffb3f08137bc20 R08: ffffffff82dc18a0 R09: 00000000000225c0
[97703.543087] R10: 0000000000000000 R11: 0000000000007eec R12: ffffffff816e4ee0
[97703.543091] R13: ffff96e06f6a5c00 R14: ffff96e077ba7700 R15: ffffffff812ab0c0
[97703.543097] FS: 0000000000000000(0000) GS:ffff96e077a80000(0000) knlGS:0000000000000000
[97703.543101] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[97703.543104] CR2: 00007f8cd135b280 CR3: 00000001e860e003 CR4: 00000000003606e0
[97703.543109] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[97703.543112] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[97703.543115] Call Trace:
[97703.543129] __flush_work+0xbd/0x1e0
[97703.543137] ? __cancel_work_timer+0x136/0x1b0
[97703.543145] ? pwq_dec_nr_in_flight+0x49/0xa0
[97703.543154] __cancel_work_timer+0x136/0x1b0
[97703.543175] ? mlxsw_reg_trans_bulk_wait+0x145/0x400 [mlxsw_core]
[97703.543184] cancel_work_sync+0x10/0x20
[97703.543191] rhashtable_free_and_destroy+0x23/0x140
[97703.543198] rhashtable_destroy+0xd/0x10
[97703.543254] mlxsw_sp_fib_destroy+0xb1/0xf0 [mlxsw_spectrum]
[97703.543310] mlxsw_sp_vr_put+0xa8/0xc0 [mlxsw_spectrum]
[97703.543364] mlxsw_sp_fib_node_put+0xbf/0x140 [mlxsw_spectrum]
[97703.543418] ? mlxsw_sp_fib6_entry_destroy+0xe8/0x110 [mlxsw_spectrum]
[97703.543475] mlxsw_sp_router_fib6_event_work+0x6cd/0x7f0 [mlxsw_spectrum]
[97703.543484] process_one_work+0x1fd/0x400
[97703.543493] worker_thread+0x34/0x410
[97703.543500] kthread+0x121/0x140
[97703.543507] ? process_one_work+0x400/0x400
[97703.543512] ? kthread_park+0x90/0x90
[97703.543523] ret_from_fork+0x35/0x40
Fixes: a3832b31898f ("mlxsw: core: Create an ordered workqueue for FIB offload")
Signed-off-by: Ido Schimmel <[email protected]>
Reported-by: Semion Lisyansky <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The EMAD workqueue is used to handle retransmission of EMAD packets that
contain configuration data for the device's firmware.
Given the workers need to allocate these packets and that the code is
not called as part of memory reclaim path, remove the WQ_MEM_RECLAIM
flag.
Fixes: d965465b60ba ("mlxsw: core: Fix possible deadlock")
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The driver cannot guarantee in the prepare phase that it will be able to
write an MDB entry to the device. In case the driver returned success
during the prepare phase, but then failed to add the entry in the commit
phase, a WARNING [1] will be generated by the switchdev core.
Fix this by doing the work in the prepare phase instead.
[1]
[ 358.544486] swp12s0: Commit of object (id=2) failed.
[ 358.550061] WARNING: CPU: 0 PID: 30 at net/switchdev/switchdev.c:281 switchdev_port_obj_add_now+0x9b/0xe0
[ 358.560754] CPU: 0 PID: 30 Comm: kworker/0:1 Not tainted 5.0.0-custom-13382-gf2449babf221 #1350
[ 358.570472] Hardware name: Mellanox Technologies Ltd. MSN2100-CB2FO/SA001017, BIOS 5.6.5 06/07/2016
[ 358.580582] Workqueue: events switchdev_deferred_process_work
[ 358.587001] RIP: 0010:switchdev_port_obj_add_now+0x9b/0xe0
...
[ 358.614109] RSP: 0018:ffffa6b900d6fe18 EFLAGS: 00010286
[ 358.619943] RAX: 0000000000000000 RBX: ffff8b00797ff000 RCX: 0000000000000000
[ 358.627912] RDX: ffff8b00b7a1d4c0 RSI: ffff8b00b7a152e8 RDI: ffff8b00b7a152e8
[ 358.635881] RBP: ffff8b005c3f5bc0 R08: 000000000000022b R09: 0000000000000000
[ 358.643850] R10: 0000000000000000 R11: ffffa6b900d6fcc8 R12: 0000000000000000
[ 358.651819] R13: dead000000000100 R14: ffff8b00b65a23c0 R15: 0ffff8b00b7a2200
[ 358.659790] FS: 0000000000000000(0000) GS:ffff8b00b7a00000(0000) knlGS:0000000000000000
[ 358.668820] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 358.675228] CR2: 00007f00aad90de0 CR3: 00000001ca80d000 CR4: 00000000001006f0
[ 358.683188] Call Trace:
[ 358.685918] switchdev_port_obj_add_deferred+0x13/0x60
[ 358.691655] switchdev_deferred_process+0x6b/0xf0
[ 358.696907] switchdev_deferred_process_work+0xa/0x10
[ 358.702548] process_one_work+0x1f5/0x3f0
[ 358.707022] worker_thread+0x28/0x3c0
[ 358.711099] ? process_one_work+0x3f0/0x3f0
[ 358.715768] kthread+0x10d/0x130
[ 358.719369] ? __kthread_create_on_node+0x180/0x180
[ 358.724815] ret_from_fork+0x35/0x40
Fixes: 3a49b4fde2a1 ("mlxsw: Adding layer 2 multicast support")
Signed-off-by: Ido Schimmel <[email protected]>
Reported-by: Alex Kushnarov <[email protected]>
Tested-by: Alex Kushnarov <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Create a new prio in the FDB, it will be used when inserting steering rules
into the FDB from the RDMA side. We create a new PRIO so rules from the
net side and rules from the RDMA side won't be inserted to the same PRIO,
each side has it's own sandbox to play in.
Signed-off-by: Mark Bloch <[email protected]>
Reviewed-by: Maor Gottlieb <[email protected]>
Signed-off-by: Leon Romanovsky <[email protected]>
|
|
When creating the FDB prios, use the enum values already defined and not
the hardcoded values.
Signed-off-by: Mark Bloch <[email protected]>
Reviewed-by: Maor Gottlieb <[email protected]>
Signed-off-by: Leon Romanovsky <[email protected]>
|
|
Although XOR hash function can perform very well on some special use
cases, to align with all drivers, mlx5 driver should use Toeplitz hash
by default.
Toeplitz is more stable for the general use case and it is more standard
and reliable.
On top of that, since XOR (MLX5_RX_HASH_FN_INVERTED_XOR8) gives only a
repeated 8 bits pattern. When used for udp tunneling RSS source port
manipulation it results in fixed source port, which will cause bad RSS
spread.
Fixes: 2be6967cdbc9 ("net/mlx5e: Support ETH_RSS_HASH_XOR")
Signed-off-by: Konstantin Khlebnikov <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Moshe Shemesh <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
This reverts commit b820e6fb0978f9c2ac438c199d2bb2f35950e9c9.
Prior the commit we are reverting, checksum unnecessary was only set when
both the L3 OK and L4 OK bits are set on the CQE. This caused packets
of IP protocols such as SCTP which are not dealt by the current HW L4
parser (hence the L4 OK bit is not set, but the L4 header type none bit
is set) to go through the checksum none code, where currently we wrongly
report checksum unnecessary for them, a regression. Fix this by a revert.
Note that on our usual track we report checksum complete, so the revert
isn't expected to have any notable performance impact. Also, when we are
not on the checksum complete track, the L4 protocols for which we report
checksum none are not high performance ones, we will still report
checksum unnecessary for UDP/TCP.
Fixes: b820e6fb0978 ("net/mlx5e: Enable reporting checksum unnecessary also for L3 packets")
Signed-off-by: Or Gerlitz <[email protected]>
Reported-by: Avi Urman <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
TC encap offload is supported only for the physical uplink
representor. Fail for non uplink representor.
Fixes: 3e621b19b0bb ("net/mlx5e: Support TC encapsulation offloads with upper devices")
Signed-off-by: Dmytro Linkin <[email protected]>
Reviewed-by: Eli Britstein <[email protected]>
Reviewed-by: Vlad Buslov <[email protected]>
Reviewed-by: Roi Dayan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
In the two places is_last_ethertype_ip is being called, the caller will
be looking inside the ip header, to be safe, add ip{4,6} header sanity
check. And return true only on valid ip headers, i.e: the whole header
is contained in the linear part of the skb.
Note: Such situation is very rare and hard to reproduce, since mlx5e
allocates a large enough headroom to contain the largest header one can
imagine.
Fixes: fe1dc069990c ("net/mlx5e: don't set CHECKSUM_COMPLETE on SCTP packets")
Reported-by: Cong Wang <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
When an ethernet frame with ip payload is padded, the padding octets are
not covered by the hardware checksum.
Prior to the cited commit, skb checksum was forced to be CHECKSUM_NONE
when padding is detected. After it, the kernel will try to trim the
padding bytes and subtract their checksum from skb->csum.
In this patch we fixup skb->csum for any ip packet with tail padding of
any size, if any padding found.
FCS case is just one special case of this general purpose patch, hence,
it is removed.
Fixes: 88078d98d1bb ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends"),
Cc: Eric Dumazet <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
XDP programs might change packets data contents which will make the
reported skb checksum (checksum complete) invalid.
When XDP programs are loaded/unloaded set/clear rx RQs
MLX5E_RQ_STATE_NO_CSUM_COMPLETE flag.
Fixes: 86994156c736 ("net/mlx5e: XDP fast RX drop bpf programs support")
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
When requested to recover from error, the tx reporter might open new
channels and close the existing ones. Use safe channels switch flow in
order to guarantee opened channels at the end of the recover flow.
For this purpose, define mlx5e_safe_reopen_channels function and use it
within those flows.
Fixes: de8650a82071 ("net/mlx5e: Add tx reporter support")
Signed-off-by: Eran Ben Elisha <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Skip recover operation if interface is in down state as TX objects are
not open. This fixes a bug were the recover flow re-opened TX objects
which were not opened before, leading to a possible memory leak at
driver unload.
Fixes: de8650a82071 ("net/mlx5e: Add tx reporter support")
Signed-off-by: Eran Ben Elisha <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Flow is kfreed on mlx5_fpga_tls_del_flow but kept in the idr data
structure, this is risky and can cause use-after-free, since the
idr_remove is delayed until tls_send_teardown_cmd completion.
Instead of delaying idr_remove, in this patch we do it on
mlx5_fpga_tls_del_flow, before actually kfree(flow).
Added synchronize_rcu before kfree(flow)
Fixes: ab412e1dd7db ("net/mlx5: Accel, add TLS rx offload routines")
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
To avoid use-after-free, hold the rcu read lock until we are done copying
flow data into the command buffer.
Fixes: ab412e1dd7db ("net/mlx5: Accel, add TLS rx offload routines")
Reported-by: Eric Dumazet <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
|