When a call is disconnected, clear the call's pointer to the connection and
release the associated ref on that connection. This means that the call no
longer pins the connection and the connection can be discarded even before
the call is.
As the code currently stands, the call struct is effectively pinned by
userspace until userspace has enacted a recvmsg() to retrieve the final
call state as sk_buffs on the receive queue pin the call to which they're
related because:
(1) The rxrpc_call struct contains the userspace ID that recvmsg() has to
include in the control message buffer to indicate which call is being
referred to. This ID must remain valid until the terminal packet is
completely read and must be invalidated immediately at that point as
userspace is entitled to immediately reuse it.
(2) The final ACK to the reply to a client call isn't sent until the last
data packet is entirely read (it's probably worth altering this in
future to be send the ACK as soon as all the data has been received).
This change requires a bit of rearrangement to make sure that the call
isn't going to try and access the connection again after protocol
completion:
(1) Delete the error link earlier when we're releasing the call. Possibly
network errors should be distributed via connections at the cost of
adding in an access to the rxrpc_connection struct.
(2) Remove the call from the connection's call tree before disconnecting
the call. The call tree needs to be removed anyway and incoming
packets delivered by channel pointer instead.
(3) The release call event should be considered last after all other
events have been processed so that we don't need access to the
connection again.
(4) Move the channel_lock taking from rxrpc_release_call() to
rxrpc_disconnect_call() where it will be required in future.
Signed-off-by: David Howells <dhowells@redhat.com>
If rxrpc_connect_call() fails during the creation of a client connection,
there are two bugs that we can hit that need fixing:
(1) The call state should be moved to RXRPC_CALL_DEAD before the call
cleanup phase is invoked. If not, this can cause an assertion failure
later.
(2) call->link should be reinitialised after being deleted in
rxrpc_new_client_call() - which otherwise leads to a failure later
when the call cleanup attempts to delete the link again.
Signed-off-by: David Howells <dhowells@redhat.com>
Rather than calling rxrpc_get_connection() manually before calling
rxrpc_queue_conn(), do it inside the queue wrapper.
This allows us to do some important fixes:
(1) If the usage count is 0, do nothing. This prevents connections from
being reanimated once they're dead.
(2) If rxrpc_queue_work() fails because the work item is already queued,
retract the usage count increment which would otherwise be lost.
(3) Don't take a ref on the connection in the work function. By passing
the ref through the work item, this is unnecessary. Doing it in the
work function is too late anyway. Previously, connection-directed
packets held a ref on the connection, but that's not really the best
idea.
And another useful changes:
(*) Don't need to take a refcount on the connection in the data_ready
handler unless we invoke the connection's work item. We're using RCU
there so that's otherwise redundant.
Signed-off-by: David Howells <dhowells@redhat.com>
Check that the client conns cache is empty before module removal and bug if
not, listing any offending connections that are still present. Unfortunately,
if there are connections still around, then the transport socket is still
unexpectedly open and active, so we can't just unallocate the connections.
Signed-off-by: David Howells <dhowells@redhat.com>
Turn the connection event and state #define lists into enums and move
outside of the struct definition.
Whilst we're at it, change _SERVER to _SERVICE in those identifiers and add
EV_ into the event name to distinguish them from flags and states.
Also add a symbol indicating the number of states and use that in the state
text array.
Signed-off-by: David Howells <dhowells@redhat.com>
Provide queueing helper functions so that the queueing of local and
connection objects can be fixed later.
The issue is that a ref on the object needs to be passed to the work queue,
but the act of queueing the object may fail because the object is already
queued. Testing the queuedness of an object before hand doesn't work
because there can be a race with someone else trying to queue it. What
will have to be done is to adjust the refcount depending on the result of
the queue operation.
Signed-off-by: David Howells <dhowells@redhat.com>
rxkad uses stack memory in SG lists which would not work if stacks were
allocated from vmalloc memory. In fact, in most cases this isn't even
necessary as the stack memory ends up getting copied over to kmalloc
memory.
This patch eliminates all the unnecessary stack memory uses by supplying
the final destination directly to the crypto API. In two instances where a
temporary buffer is actually needed we also switch use a scratch area in
the rxrpc_call struct (only one DATA packet will be being secured or
verified at a time).
Finally there is no need to split a split-page buffer into two SG entries
so code dealing with that has been removed.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: David Howells <dhowells@redhat.com>
When looking up a client connection to which to route a packet, we need to
check that the packet came from the correct source so that a peer can't try
to muck around with another peer's connection.
Signed-off-by: David Howells <dhowells@redhat.com>
When a jumbo packet is being split up and processed, the crypto checksum
for each split-out packet is in the jumbo header and needs placing in the
reconstructed packet header.
When the code was changed to keep the stored copy of the packet header in
host byte order, this reconstruction was missed.
Found with sparse with CF=-D__CHECK_ENDIAN__:
../net/rxrpc/input.c:479:33: warning: incorrect type in assignment (different base types)
../net/rxrpc/input.c:479:33: expected unsigned short [unsigned] [usertype] _rsvd
../net/rxrpc/input.c:479:33: got restricted __be16 [addressable] [usertype] _rsvd
Fixes: 0d12f8a402 ("rxrpc: Keep the skb private record of the Rx header in host byte order")
Signed-off-by: David Howells <dhowells@redhat.com>
Russell King says:
====================
Initial SFP support patches
Please review and merge this initial patch set, which is part of a
larger set previously posted adding SFP support to phy and mvneta.
This initial set are focused on cleaning up and reorganising the
fixed-phy code to allow the core software-phy code to be re-used.
These are based on net-next.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no prevention of a concurrent call to both fixed_mdio_read()
and fixed_phy_update_state(), which can result in the state being
modified while it's being inspected. Fix this by using a seqcount
to detect modifications, and memcpy()ing the state.
We remain slightly naughty here, calling link_update() and updating
the link status within the read-side loop - which would need rework
of the design to change.
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Generate software phy registers as and when requested, rather than
duplicating the state in fixed_phy. This allows us to eliminate
the duplicate storage of of the same data, which is only different
in format.
As fixed_phy_update_regs() no longer updates register state, rename
it to fixed_phy_update().
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Separate out the generation of MII registers from the state validation.
This allows us to simplify the error handing in fixed_phy() by allowing
earlier error detection.
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert the swphy register generation to tabular form which allows us
to eliminate multiple switch() statements. This results in a smaller
object code size, more efficient, and easier to add support for faster
speeds.
Before:
Idx Name Size VMA LMA File off Algn
0 .text 00000164 00000000 00000000 00000034 2**2
text data bss dec hex filename
388 0 0 388 184 swphy.o
After:
Idx Name Size VMA LMA File off Algn
0 .text 000000fc 00000000 00000000 00000034 2**2
5 .rodata 00000028 00000000 00000000 00000138 2**2
text data bss dec hex filename
324 0 0 324 144 swphy.o
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the fixed_phy MII register generation to a library to allow other
software phy implementations to use this code.
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
-----BEGIN PGP SIGNATURE-----
iQEcBAABCgAGBQJXa6rGAAoJED07qiWsqSVqzwsH/0g/blikQuNNAB7LzXGTEMn8
ipWCODaYxgWE/V/1A0JjsKpxB1ZL3HE0/xg/OmZ10SRNlz5LJTLkoYZa7xYmyxtT
a89mT9vdnWxwfwbeIhSSjG9W9g44vqrhMIPxffIu5yqSqdIJak0HIiuQizJ0/xjx
xQVx4AykWKQ3u0/Tiyz3ez5yhyMMNEmHxKyDWpgrR6+zXlofwP/Em3NFPwk9gh32
ECAagOUOEvdjXeRs1Yn/CV0FC7Wgs4Hzr048JJ5wOteawBLr+sPOnuWZtDhNrwDK
Eceh7aEeAiFfIW+jCMEe5qrpFHi8hqyqBo4L9TDYsE7T/Z6BPyh+5OUh8+5dtaw=
=TZew
-----END PGP SIGNATURE-----
Merge tag 'linux-can-next-for-4.8-20160623' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next
Marc Kleine-Budde says:
====================
pull-request: can-next 2016-06-17
this is a pull request of 4 patches for net-next/master.
Arnd Bergmann's patch fixes a regresseion in af_can introduced in
linux-can-next-for-4.8-20160617. There are two patches by Ramesh
Shanmugasundaram, which add CAN-2.0 support to the rcar_canfd driver.
And a patch by Ed Spiridonov that adds better error diagnoses messages
to the Ed Spiridonov driver.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace calls to kmalloc followed by a memcpy with a direct call to
kmemdup.
The Coccinelle semantic patch used to make this change is as follows:
@@
expression from,to,size,flag;
statement S;
@@
- to = \(kmalloc\|kzalloc\)(size,flag);
+ to = kmemdup(from,size,flag);
if (to==NULL || ...) S
- memcpy(to, from, size);
Signed-off-by: Amitoj Kaur Chawla <amitoj1606@gmail.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
trivial fixes to spelling mistakes of the words "excessive collisions"
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
trivial fixes to spelling mistakes of the word "descriptors"
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Saeed Mahameed says:
====================
Mellanox 100G mlx5e Ethernet extensions
This series includes multiple features extensions for mlx5 Ethernet netdevice driver.
Namely, TX Rate limiting, RX interrupt moderation, ethtool settings.
TX Rate limiting:
- ConnectX-4 rate limiting infrastructure
- Set max rate NDO support
RX interrupt moderation:
- CQE based coalescing option (controlled via priv flags)
- Adaptive RX coalescing
ethtool settings:
- priv flags callbacks
- Support new ksettings API
- Add 50G missing link mode
- Support auto negotiation on/off
Applied on top: 0e9390ebf1 ("Merge branch 'mlxsw-next'")
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Previous to this patch auto negotiation was reported off although it was
on by default in hardware. This patch reports the correct information to
ethtool and allows the user to toggle it on/off.
Added another parameter to set port proto function in order to pass
the auto negotiation field to the hardware.
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use new get/set link ksettings and remove get/set settings legacy
callbacks.
This allows us to use bitmasks longer than 32 bit for supported and
advertised link modes and use modes that were previously not supported.
Signed-off-by: Gal Pressman <galp@mellanox.com>
CC: Ben Hutchings <bwh@kernel.org>
CC: David Decotigny <decot@googlers.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add MLX5E_50GBASE_SR2 as ETHTOOL_LINK_MODE_50000baseSR2_Full_BIT.
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Cc: Ben Hutchings <bwh@kernel.org>
Cc: David Decotigny <decot@googlers.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add ETHTOOL_LINK_MODE_50000baseSR2_Full_BIT bit.
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Cc: Ben Hutchings <bwh@kernel.org>
Cc: David Decotigny <decot@googlers.com>
Acked-By: David Decotigny <decot@googlers.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a dedicated function to toggle port link. It should be called only
after setting a port register.
Toggle will set port link to down and bring it back up in case that it's
admin status was up.
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In this mode the moderation timer will restart upon
new completion (CQE) generation rather than upon interrupt
generation.
The outcome is that for bursty traffic the period timer will never
expire and thus only the moderation frames counter will dictate
interrupt generation, thus the interrupt rate will be relative
to the incoming packets size.
If the burst seizes for "moderation period" time then an interrupt
will be issued immediately.
CQE based moderation is off by default and can be controlled
via ethtool set_priv_flags.
Performance tested on ConnectX4-Lx 50G.
Less packet loss in netperf UDP and TCP tests, with no bw degradation,
for both single and multi streams, with message sizes of
64, 1024, 1472 and 32768 byte.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Gil Rockah <gilr@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce an infrastructure for getting/setting private net device
flags.
Currently a 'nop' priv flag is added, following patches will override
the flag will actual feature specific flags.
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement set_maxrate ndo.
Use the rate index from the hardware table to attach to channel SQ/TXQ.
In case of failure to configure new rate, the queue remains with
unlimited rate.
We save the configuration on priv structure and apply it each time
Send Queues are being reinitialized (after open/close) operations.
Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Configuring and managing HW rate limit tables.
The HW holds a table of rate limits, each rate is
associated with an index in that table.
Later a Send Queue uses this index to set the rate limit.
Multiple Send Queues can have the same rate limit, which is
represented by a single entry in this table.
Even though a rate can be shared, each queue is being rate
limited independently of others.
The SW shadow of this table holds the rate itself,
the index in the HW table and the refcount (number of queues)
working with this rate.
The exported functions are mlx5_rl_add_rate and mlx5_rl_remove_rate.
Number of different rates and their values are derived
from HW capabilities.
Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sathya Perla says:
====================
be2net: patch set
Hi Dave, pls consider commiting the following patches to the net-next tree.
Thanks!
Patch 1 replaces the be_max_eqs() macro with two new macros called
be_max_nic_eqs() and be_max_func_eqs() to clear confusion in that part
of the code.
Patch 2 adds support to configure asymmetric number of rx/tx queues via
ethtool set-channels option.
Patch 3 disables EVB when VFs are not enabled on a BE3 SR-IOV config to
avoid the broadcast echo problem.
Patch 4 updates copyright markings in be2net src files
Patch 5 updates the be2net maintainers' list
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes Padmanabh's name from the maintainers list as he's no
longer with the company. It also adds the driver name on the headline to
make it easy to lookup the maintainers list by the driver name.
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch updates year and company name in the copyright markings in the
be2net source files.
Signed-off-by: Somnath Kotur <somnath.kotur@emulex.com>
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On SR-IOV profiles, when the user connects a Linux Bridge or OVS to a BE3
vport, they suffer the "broadcast/multicast echo" problem. BE3 EVB echoes
broadcast and multicast packets back to PF's vport confusing the
Linux bridge. BE3 relies on the src-mac addr being programmed on the
interface to avoid sending back an echo of a broadcast or multicast packet
on a vPort. When a Linux bridge is connected to a BE3, the mac-addr of the
VM behind the bridge doesn't get configured on the vPort and so echo
cancellation doesn't work.
This patch worksaround this problem by disabling the EVB initially
and re-enabling it *only* when SR-IOV is enabled by the user. For the
driver fix to work, the BE3 FW version must be >= 11.1.84.0.
Signed-off-by: Somnath Kotur <somnath.kotur@emulex.com>
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
be2net so far supported creation of RX/TX queues only in pairs.
On configs where rx and tx queue counts are different, creation of only
the lesser number of queues has been supported.
This patch now allows a combination of RX/TX-only channels along with
combined channels. N TX-queues and M RX-queues can be created with the
following cmds:
ethtool -L ethX combined N rx M-N (when N < M)
ethtool -L ethX combined M tx N-M (when M < N)
Setting both RX-only and TX-only channels is still not supported.
It is mandatory to create atleast one combined channel.
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The EQs available on a function are shared between NIC and RoCE.
The be_max_eqs() macro was so far being used to refer to the max number of
EQs available for NIC. This has caused some confusion in the code. To fix
this confusion this patch introduces a new macro called be_max_nic_eqs()
to refer to the max number of EQs avialable for NIC only and renames
be_max_eqs() to be_max_func_eqs().
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Andy Duan says:
====================
net: fec: add new type device
Different i.MX SOC FEC support different features like :
- i.MX6Q/DL FEC does not support AVB and interrupt coalesc
- i.MX6SX/i.MX7D supports AVB and interrupt coalesc
- i.MX6UL/ULL does not support AVB, but support interrupt coalesc
Then, add new quirk flag to judge the supported features, and add new
type device for i.MX6UL.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
i.MX6UL is a member in i.MX series family, the SOC FEC inherits from
i.MX6SX but removes some IP features, lets define a new type for fec
device.
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Different i.MX SOC FEC support different features like :
- i.MX6Q/DL FEC does not support AVB and interrupt coalesc
- i.MX6SX/i.MX7D supports AVB and interrupt coalesc
- i.MX6UL/ULL does not support AVB, but support interrupt coalesc
So, add new quirk flag to judge the supported features.
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=HxJc
-----END PGP SIGNATURE-----
Merge tag 'rxrpc-rewrite-20160622-2' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
David Howells says:
====================
rxrpc: Get rid of conn bundle and transport structs
Here's the next part of the AF_RXRPC rewrite. The primary purpose of this
set is to get rid of the rxrpc_conn_bundle and rxrpc_transport structs.
This simplifies things for future development of the connection handling.
To this end, the following significant changes are made:
(1) The rxrpc_connection struct is given pointers to the local and peer
endpoints, inside the rxrpc_conn_parameters struct. Pointers to the
transport's copy of these pointers are then redirected to the
connection struct.
(2) Exclusive connection handling is fixed. Exclusive connections should
do just one call and then be retired. They are used in security
negotiations and, I believe, the idea is to avoid reuse of negotiated
security contexts.
The current code is doing a single connection per socket and doing all
the calls over that. With this change it gets a new connection for
each call made.
(3) A new sendmsg() control message marker is added to make individual
calls operate over exclusive connections. This should be used in
future in preference to the sockopt that marks a socket as "exclusive
connection".
(4) IDs for client connections initiated by a machine are now allocated
from a global pool using the IDR facility and are unique across all
client connections, no matter their destination. The IDR facility is
then used to look up a connection on the connection ID alone. Other
parameters are then verified afterwards.
Note that the IDR facility may use a lot of memory if the IDs it holds
are widely scattered. Given this, in a future commit, client
connections will be retired if they are more than a certain distance
from the last ID allocated.
The client epoch is advanced by 1 each time the client ID counter
wraps. Connections outside the current epoch will also be retired in
a future commit.
(5) The connection bundle concept is removed and the client connection
tree is moved into the local endpoint. The queue for waiting for a
call channel is moved to the rxrpc_connection struct as there can only
be one connection for any particular key going to any particular peer
now.
(6) The rxrpc_transport struct is removed and the service connection tree
is moved into the peer struct.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add constants and callback functions for the dwmac on rk3228/rk3229 socs.
As can be seen, the base structure is the same, only registers and the
bits in them moved slightly.
Signed-off-by: Xing Zheng <zhengxing@rock-chips.com>
Reviewed-by: Heiko Stuebner <heiko@sntech.de>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet says:
====================
net_sched: bulk dequeue and deferred drops
First patch adds an additional parameter to ->enqueue() qdisc method
so that drops can be done outside of critical section
(after locks are released).
Then fq_codel can have a small optimization to reduce number of cache
lines misses during a drop event
(possibly accumulating hundreds of packets to be freed).
A small htb change exports the backlog in class dumps.
Final patch adds bulk dequeue to qdiscs that were lacking this feature.
This series brings a nice qdisc performance increase (more than 80 %
in some cases).
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
When qdisc bulk dequeue was added in linux-3.18 (commit
5772e9a346 "qdisc: bulk dequeue support for qdiscs
with TCQ_F_ONETXQUEUE"), it was constrained to some
specific qdiscs.
With some extra care, we can extend this to all qdiscs,
so that typical traffic shaping solutions can benefit from
small batches (8 packets in this patch).
For example, HTB is often used on some multi queue device.
And bonding/team are multi queue devices...
Idea is to bulk-dequeue packets mapping to the same transmit queue.
This brings between 35 and 80 % performance increase in HTB setup
under pressure on a bonding setup :
1) NUMA node contention : 610,000 pps -> 1,110,000 pps
2) No node contention : 1,380,000 pps -> 1,930,000 pps
Now we should work to add batches on the enqueue() side ;)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.r.fastabend@intel.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Florian Westphal <fw@strlen.de>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We already get child qdisc qlen, we also can get its backlog
so that class dumps can report it.
Also replace qstats by a single drop counter, but move it in
a separate cache line so that drops do not dirty useful cache lines.
Tested:
$ tc -s cl sh dev eth0
class htb 1:1 root leaf 3: prio 0 rate 1Gbit ceil 1Gbit burst 500000b cburst 500000b
Sent 2183346912 bytes 9021815 pkt (dropped 2340774, overlimits 0 requeues 0)
rate 1001Mbit 517543pps backlog 120758b 499p requeues 0
lended: 9021770 borrowed: 0 giants: 0
tokens: 9 ctokens: 9
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now we defer skb drops, it makes sense to keep a copy
of skb->truesize in struct codel_skb_cb to avoid one
cache line miss per dropped skb in fq_codel_drop(),
to reduce latencies a bit further.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Qdisc performance suffers when packets are dropped at enqueue()
time because drops (kfree_skb()) are done while qdisc lock is held,
delaying a dequeue() draining the queue.
Nominal throughput can be reduced by 50 % when this happens,
at a time we would like the dequeue() to proceed as fast as possible.
Even FQ is vulnerable to this problem, while one of FQ goals was
to provide some flow isolation.
This patch adds a 'struct sk_buff **to_free' parameter to all
qdisc->enqueue(), and in qdisc_drop() helper.
I measured a performance increase of up to 12 %, but this patch
is a prereq so that future batches in enqueue() can fly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Raghu Vatsavayi says:
====================
liquidio: updates and bug fixes
Please consider following patch series for liquidio bug fixes
and updates on top of net-next. Following patches should be
applied in the following order as some of them depend on
earlier patches in the series.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds support for ddr_timeout during device init.
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for private flags for the driver.
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
