During fm10k_io_error_detected we were clearing the interrupt scheme
before we freed the MBX IRQ. This causes a kernel panic because the MBX
IRQ are assigned after MSI-X initialization. Clearing the interrupt
scheme results in removing the MSI-X entry table. Fix this by freeing
the MBX IRQ before we clear the interrupt scheme, as we do elsewhere in
the driver.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
fm10k_stop_hw_generic calls fm10k_disable_queues_generic, which may
return an error code indicating that the queues were not stopped within
the time limit. Notify the user by displaying a message in the kernel
message ring, in a similar way to how we notify the user when reset_hw
fails. There isn't much we can do to recover from this error, so
currently nothing else is done.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In fm10k_set_num_queues, we previously assigned the base template. This
would always be overwritten by either fm10k_set_qos_queues or
fm10k_set_rss_queues. In either case, we don't need the base values, so
we can just remove them.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
According to the C standard dereferencing a variable before it is
checked invokes undefined behavior, and thus compilers are free to
assume the check for NULL isn't necessary. Prevent this by re-ordering
the NULL check of msix_entries in fm10k_free_mbx_irq.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cleanup the remaining instances of using memcpy() instead of the preferred
ether_addr_copy().
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
We don't need to crash the kernel in this instance so just warn about the
condition and play on.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Use BIT() macro instead.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The semantic patch that makes this change is available
in scripts/coccinelle/misc/compare_const_fl.cocci.
More information about semantic patching is available at
http://coccinelle.lip6.fr/
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Minor correction in the comment to reflect the correct function name
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
If the driver happens to read a register during the time in which the
device is undergoing reset, it will receive a value of 0xdeadbeef
instead of a valid value. Unfortunately, the driver may misinterpret
this as a valid value, especially if it's just looking for individual
bits.
Add an explicit check for this value when we are looking for admin queue
errors, and trigger reset recovery if we find it.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
There's no real error in an unknown event from the Firmware, we're just
posting a useful FYI notice, so this patch simply removes the "Error" word.
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Clear the VFLR bit after reset processing, instead of before. This
prevents double resets on VF init.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Notify VFs in the reset interrupt handler, instead of the actual
reset initiation code. This allows the VFs to get properly notified for
all resets, including resets initiated by different PFs on the same
physical device.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In some error scenarios, we may find ourselves trying to remove a
non-existent timer or worktask. This causes the kernel some bit
of consternation, so don't do it.
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Jiri Pirko says:
====================
mlxsw: small driver update, including switchdev doc update
Ido Schimmel (3):
mlxsw: spectrum: Reduce number of supported 802.1D bridges
switchdev: Use switch ID in suggested udev rule
mlxsw: spectrum: Add support for physical port names
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Export to userspace the front panel name of the port, so that udev can
rename the ports accordingly. The convention suggested by switchdev
documentation is used:
1) Non-split: pX
2) Split: pXsY
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since there can be multiple switch ASICs on the same system we should
use the switch ID in order to differentiate between them and set the
switch name (e.g. swX) accordingly.
Also, replace the order of the "Switch ID" and "Port Netdev Naming"
sections following the above change.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Resources allocated for these bridges at init time cannot be later used
for other purposes. While current number is supported by the device,
it's mostly theoretical with regards to any real use case, which leads
to poor utilization of device's resources. Solve that by reducing the
number.
The long term plan is to make this value (along with others) user
configurable via devlink and write it to NVRAM, so that it can be used
during the next init. Until then we must hardcode such values.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When dropping into debug mode in a failed probe, make sure that
the AdminQ is left alive for possible hand debug of driver and
firmware states.
Move the mutex_init calls earlier in probe so that if init fails,
the admin queue interface is still available for debugging purposes.
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When cleaning up the interrupt handling, clean up the IRQs only if
we actually got them set up. There are a couple of error recovery
paths that were violating this and causing the kernel a bit of
indigestion.
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Reviewed-by: Williams, Mitch A <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The i40e_common.c typically uses i40e_status as a return code,
but got missed this one case.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When updating a VSI, save off the number of allocated and unallocated
VSIs as we do when adding a VSI.
Signed-off-by: Kevin Scott <kevin.c.scott@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch removes the duplicate definition of I40E_MAX_USER_PRIORITY
in i40e.h that is not needed.
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Simple cast to fix a sparse warning.
Fixes: commit 5453205cd0 ("i40e/i40evf: Enable support for
SKB_GSO_UDP_TUNNEL_CSUM")
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch enables bulk Tx clean for skbs. In order to enable it we need
to pass the napi_budget value as that is used to determine if we are truly
running in NAPI mode or if we are simply calling the routine from netpoll
with a budget of 0. In order to avoid adding too many more variables I
thought it best to pass the VSI directly in a fashion similar to what we do
on igb and ixgbe with the q_vector.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In the polling routines for i40e and i40evf we were using bitwise operators
to avoid the side effects of the logical operators, specifically the fact
that if the first case is true with "||" we skip the second case, or if it
is false with "&&" we skip the second case. This fixes an earlier patch
that converted the bitwise operators over to the logical operators and
instead replaces the entire thing with just an if statement since it should
be more readable what we are trying to do this way.
Fixes: 1a36d7fadd ("i40e/i40evf: use logical operators, not bitwise")
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The only error case is when the malloc fails, in which case the clean up
loop does nothing at all, so remove it
Signed-off-by: Alan Cox <alan@linux.intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
From what I can tell the practical limitation on the size of the Tx data
buffer is the fact that the Tx descriptor is limited to 14 bits. As such
we cannot use 16K as is typically used on the other Intel drivers. However
artificially limiting ourselves to 8K can be expensive as this means that
we will consume up to 10 descriptors (1 context, 1 for header, and 9 for
payload, non-8K aligned) in a single send.
I propose that we can reduce this by increasing the maximum data for a 4K
aligned block to 12K. We can reduce the descriptors used for a 32K aligned
block by 1 by increasing the size like this. In addition we still have the
4K - 1 of space that is still unused. We can use this as a bit of extra
padding when dealing with data that is not aligned to 4K.
By aligning the descriptors after the first to 4K we can improve the
efficiency of PCIe accesses as we can avoid using byte enables and can fetch
full TLP transactions after the first fetch of the buffer. This helps to
improve PCIe efficiency. Below is the results of testing before and after
with this patch:
Recv Send Send Utilization Service Demand
Socket Socket Message Elapsed Send Recv Send Recv
Size Size Size Time Throughput local remote local remote
bytes bytes bytes secs. 10^6bits/s % S % U us/KB us/KB
Before:
87380 16384 16384 10.00 33682.24 20.27 -1.00 0.592 -1.00
After:
87380 16384 16384 10.00 34204.08 20.54 -1.00 0.590 -1.00
So the net result of this patch is that we have a small gain in throughput
due to a reduction in overhead for putting together the frame.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Calling dev_close() causes IFF_UP to be cleared which will remove the
interfaces routes and some addresses. That's probably not what the user
intended when running the offline selftest. Besides this does not happen
if the interface is brought down before the test, so the current
behaviour is inconsistent.
Instead call the net_device_ops ndo_stop function directly and avoid
touching IFF_UP at all.
Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Eric Dumazet says:
====================
net: various udp/tcp changes
First round of patches for linux-4.7
Add a generic facility for sockets to be freed after an RCU grace
period, if they need to.
Then UDP stack is changed to no longer use SLAB_DESTROY_BY_RCU,
in order to speedup rx processing for traffic encapsulated in UDP.
It gives a 17 % speedup for normal UDP reception in stress conditions.
Then TCP listeners are changed to use SOCK_RCU_FREE as well
to avoid touching sk_refcnt in synflood case :
I got up to 30 % performance increase for a mono listener.
Then three patches add SK_MEMINFO_DROPS to sock_diag
and add per socket rx drops accounting to TCP.
Last patch adds rate limiting on ACK sent on behalf of SYN_RECV
to better resist to SYNFLOOD targeting one or few flows.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Attackers like to use SYNFLOOD targeting one 5-tuple, as they
hit a single RX queue (and cpu) on the victim.
If they use random sequence numbers in their SYN, we detect
they do not match the expected window and send back an ACK.
This patch adds a rate limitation, so that the effect of such
attacks is limited to ingress only.
We roughly double our ability to absorb such attacks.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP uses per cpu 'sockets' to send some packets :
- RST packets ( tcp_v4_send_reset()) )
- ACK packets for SYN_RECV and TIMEWAIT sockets
By setting SOCK_USE_WRITE_QUEUE flag, we tell sock_wfree()
to not call sk_write_space() since these internal sockets
do not care.
This gives a small performance improvement, merely by allowing
cpu to properly predict the sock_wfree() conditional branch,
and avoiding one atomic operation.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Goal: packets dropped by a listener are accounted for.
This adds tcp_listendrop() helper, and clears sk_drops in sk_clone_lock()
so that children do not inherit their parent drop count.
Note that we no longer increment LINUX_MIB_LISTENDROPS counter when
sending a SYNCOOKIE, since the SYN packet generated a SYNACK.
We already have a separate LINUX_MIB_SYNCOOKIESSENT
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now ss can report sk_drops, we can instruct TCP to increment
this per socket counter when it drops an incoming frame, to refine
monitoring and debugging.
Following patch takes care of listeners drops.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reporting sk_drops to user space was available for UDP
sockets using /proc interface.
Add this to sock_diag, so that we can have the same information
available to ss users, and we'll be able to add sk_drops
indications for TCP sockets as well.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a SYNFLOOD targets a non SO_REUSEPORT listener, multiple
cpus contend on sk->sk_refcnt and sk->sk_wmem_alloc changes.
By letting listeners use SOCK_RCU_FREE infrastructure,
we can relax TCP_LISTEN lookup rules and avoid touching sk_refcnt
Note that we still use SLAB_DESTROY_BY_RCU rules for other sockets,
only listeners are impacted by this change.
Peak performance under SYNFLOOD is increased by ~33% :
On my test machine, I could process 3.2 Mpps instead of 2.4 Mpps
Most consuming functions are now skb_set_owner_w() and sock_wfree()
contending on sk->sk_wmem_alloc when cooking SYNACK and freeing them.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We'll soon no longer take a refcount on listeners,
so reqsk_alloc() can not assume a listener refcount is not
zero. We need to use atomic_inc_not_zero()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RX packet processing holds rcu_read_lock(), so we can remove
pairs of rcu_read_lock()/rcu_read_unlock() in lookup functions
if inet_diag also holds rcu before calling them.
This is needed anyway as __inet_lookup_listener() and
inet6_lookup_listener() will soon no longer increment
refcount on the found listener.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since linux 2.6.29, lookups only use rcu locking.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Herbert would like not touching UDP socket refcnt for encapsulated
traffic. For this to happen, we need to use normal RCU rules, with a grace
period before freeing a socket. UDP sockets are not short lived in the
high usage case, so the added cost of call_rcu() should not be a concern.
This actually removes a lot of complexity in UDP stack.
Multicast receives no longer need to hold a bucket spinlock.
Note that ip early demux still needs to take a reference on the socket.
Same remark for functions used by xt_socket and xt_PROXY netfilter modules,
but this might be changed later.
Performance for a single UDP socket receiving flood traffic from
many RX queues/cpus.
Simple udp_rx using simple recvfrom() loop :
438 kpps instead of 374 kpps : 17 % increase of the peak rate.
v2: Addressed Willem de Bruijn feedback in multicast handling
- keep early demux break in __udp4_lib_demux_lookup()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <tom@herbertland.com>
Cc: Willem de Bruijn <willemb@google.com>
Tested-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We want a generic way to insert an RCU grace period before socket
freeing for cases where RCU_SLAB_DESTROY_BY_RCU is adding too
much overhead.
SLAB_DESTROY_BY_RCU strict rules force us to take a reference
on the socket sk_refcnt, and it is a performance problem for UDP
encapsulation, or TCP synflood behavior, as many CPUs might
attempt the atomic operations on a shared sk_refcnt
UDP sockets and TCP listeners can set SOCK_RCU_FREE so that their
lookup can use traditional RCU rules, without refcount changes.
They can set the flag only once hashed and visible by other cpus.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <tom@herbertland.com>
Tested-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
10GbE Intel Wired LAN Driver Updates 2016-04-04
This series contains updates to ixgbe and ixgbevf.
Pavel Tikhomirov fixes a typo where we were incrementing transmit stats
instead of receive stats on the receive side.
Emil updates the ixgbevf driver to use bit operations for setting and
checking the adapter state.
Chas Williams adds the new NDO trust feature check so that the VF guest
has the ability to set the unicast address of the interface, if it is a
trusted VF.
Alex cleans up the driver to that the only time we add a PF entry to the
VLVF is either for VLAN 0 or if the PF has requested a VLAN that a VF
is already using. Also adds support for generic transmit checksums,
giving the added advantage is that we can support inner checksum offloads
for tunnels and MPLS while still being able to transparently insert
VLAN tags. Lastly, changed ixgbe so that we can use the ethtool
rx-vlan-filter flag to toggle receive VLAN filtering on and off.
Mark cleans up the ixgbe driver by making all op structures that do not
change constants. Also fixed flow control for Xeon D KR backplanes, since
we cannot use auto-negotiation to determine the mode, we have to use
whatever the user configured.
Sowmini Varadhan updates ixgbe to use eth_platform_get_mac_address()
instead of the arch specific solution that was added by a previous
commit.
Don fixed an issue where it was possible that a system reset could occur
when we were holding the SWFW semaphore lock, which the next time the
driver loaded would see it incorrectly as locked.
v2: updated patch 8 of the series to include a minor flags issue where
we had lost NETIF_F_HW_TC and we were setting NETIF_F_SCTP_CRC in
two different areas, when we only needed/wanted it in one spot.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Vivien Didelot says:
====================
net: dsa: mv88e6131: HW bridging support for 6185
All packets passing through a switch of the 6185 family are currently all
directed to the CPU port. This means that port bridging is software driven.
To enable hardware bridging for this switch family, we need to implement the
port mapping operations, the FDB operations, and optionally the VLAN operations
(for 802.1Q and VLAN filtering aware systems).
However this family only has 256 FDBs indexed by 8-bit identifiers, opposed to
4096 FDBs with 12-bit identifiers for other families such as 6352. It also
doesn't have dedicated FID registers for ATU and VTU operations.
This patchset fixes these differences, and enable hardware bridging for 6185.
Changes v1 -> v2:
- Describe the different numbers of databases and prefer a feature-based logic
over the current ID/family-based logic.
====================
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
By adding support for bridge operations, FDB operations, and optionally
VLAN operations (for 802.1Q and VLAN filtering aware systems), the
switch bridges ports correctly, the CPU is able to populate the hardware
address databases, and thus hardware bridging becomes functional within
the 88E6185 family of switches.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The 88E6185 switch also has a MapDA bit in its Port Control 2 register.
When this bit is cleared, all frames are sent out to the CPU port.
Set this bit to rely on address databases (ATU) hits and direct frames
out of the correct ports, and thus allow hardware bridging.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The 6185 family of devices has only 256 address databases. Their 8-bit
FID for ATU and VTU operations are split into ATU Control and ATU/VTU
Operation registers.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Marvell switch chips have different number of address databases.
The code currently only supports models with 4096 databases. Such switch
has dedicated FID registers for ATU and VTU operations. Models with
fewer databases have their FID split in several registers.
List them all but only support models with 4096 databases at the moment.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
