This patch provides the initial phylib driver in support
of the AMD 10GbE device.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch provides the initial platform driver for the AMD
10GbE device.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch provides the documentation of the device bindings
for the AMD 10GbE platform driver.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
br_manage_promisc() incorrectly expects br_auto_port() to return only 0
or 1, while it actually returns flags, i.e., a subset of BR_AUTO_MASK.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Ruder says:
====================
miscellaneous dm9000 driver fixes
This is a collection of changes discovered while bringing a PXA270 based board
(Arcom ZEUS) with a Davicom DM9000A/B up to a more recent kernel (from 2.6.xx).
This addresses all of my earlier issues (August 2013) listed here:
http://marc.info/?l=linux-netdev&m=137598605603324&w=2
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
On the DM9000B, dm9000_msleep() is called during the dm9000_timeout()
routine. Since dm9000_timeout() holds the main spinlock through the
entire routine, mdelay() needs to be used rather than msleep().
Furthermore, the mutex_lock()/mutex_unlock() should be avoided so as to
not sleep with spinlocks held.
Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On the DM9000A/DM9000B force the initial check of the link status. The
DM9000A/B has a link status changed event and this interrupt bit isn't
always set out of reset when a cable is plugged in. This results in the
driver not seeing the cable attached link status until the cable is
removed and plugged in again.
Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since dm9000_interrupt() is already reading/clearing every set bit in
DM9000_ISR, this additional clear in dm9000_rx() (which is only called
by dm9000_interrupt()) is unnecessary and can be removed.
Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
DM9000 uses level-triggered interrupts. Some systems (PXA270) only
support edge-triggered interrupts on GPIOs. Some changes are necessary
to ensure that interrupts are not triggered while the GPIO interrupt is
masked or we will miss the interrupt forever.
* Make some helper functions called dm9000_mask_interrupts() and
dm9000_unmask_interrupts() for readability.
* dm9000_init_dm9000(): ensure that this function always leaves interrupts
masked regardless of the state when it entered the function. This is
primarily to support the situation in dm9000_open where the logic used
to go:
dm9000_open()
dm9000_init_dm9000()
unmask interrupts
request_irq()
If an interrupt occurred between unmasking the interrupt and
requesting the irq, it would be missed forever as the edge event would
never be seen by the GPIO hardware in the PXA270. This allows us to
change the logic to:
dm9000_open()
dm9000_init_dm9000()
dm9000_mask_interrupts()
request_irq()
dm9000_unmask_interrupts()
* dm9000_timeout(), dm9000_drv_resume(): Add the missing
dm9000_unmask_interrupts() now required by the change above.
* dm9000_shutdown(): Use mask helper function
* dm9000_interrupt(): Use mask/unmask helper functions
Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* Change a hard-coded 0x3 to NCR_RST | NCR_MAC_LBK in dm9000_reset
* Every single place where dm9000_init_dm9000 was ran, a dm9000_reset
was called immediately before-hand. Bring dm9000_reset into
dm9000_init_dm9000.
* The following commit updated the dm9000_probe reset routine to use NCR_RST
| NCR_MAC_LBK:
6741f40 DM9000B: driver initialization upgrade
and a later commit added a bug-fix to always reset the chip twice:
09ee9f8 dm9000: Implement full reset of DM9000 network device
Unfortunately, since the changes in 6741f40 were made by replacing the
dm9000_probe dm9000_reset with the adjusted iow(), the changes in
09ee9f8 were not incorporated into the dm9000_probe reset.
Furthermore, it bypassed the requisite reset-delay causing some boards
to get at least one "read wrong id ..." dev_err message during
dm9000_probe.
Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DM9000 supports both active high interrupts and active low interrupts.
This is configured via the attached EEPROM. In the device-tree case, make sure
that the DM9000 driver passes the correct flags to request_irq.
Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If an MPLS packet requires segmentation then use mpls_features
to determine if the software implementation should be used.
As no driver advertises MPLS GSO segmentation this will always be
the case.
I had not noticed that this was necessary before as software MPLS GSO
segmentation was already being used in my test environment. I believe that
the reason for that is the skbs in question always had fragments and the
driver I used does not advertise NETIF_F_FRAGLIST (which seems to be the
case for most drivers). Thus software segmentation was activated by
skb_gso_ok().
This introduces the overhead of an extra call to skb_network_protocol()
in the case where where CONFIG_NET_MPLS_GSO is set and
skb->ip_summed == CHECKSUM_NONE.
Thanks to Jesse Gross for prompting me to investigate this.
Signed-off-by: Simon Horman <horms@verge.net.au>
Acked-by: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
Intel Wired LAN Driver Updates 2014-06-05
This series contains updates to i40e and i40evf.
Jesse fixes an issue reported by Dave Jones where a couple of FD checks
ended up using bitwise OR where it should have been bitwise AND.
Neerav removes unused defines and macros for receive LRO. Fix the driver
from allowing the user to set a larger MTU size that the hardware was
being configured to support. Refactors send version which moves code in
two places into a small helper function.
Kamil modifies register diagnostics since register ranges can vary among
the different NVMs to avoid false test results. So now we try to identify
the full range and use it for a register test and if we fail to define the
proper register range, we will only test the first register from that
group. Then removes the check for large buffer since this was added in the
case this structure changed in the future, since the AQ definition is now
mature enough that this check is no longer necessary.
Mitch fixes i40evf driver to allocate descriptors in groups of 32 since the
hardware requires it. Also fixes a crash when the ring size changed because
it would change the count before deallocating resources, causing the driver
to either free nonexistent buffers or leak leftover buffers. Fixed the
driver to notify the VF for all types of resets so the VF can attempt a
graceful reinit.
Shannon refactors stats collection to create a unifying stats update routine
to call the various stat collection routines. Removes rx_errors and
rx_missed stats since they were removed from the chip design. Added
missing VSI statistics that the hardware offers but are not apart of the
standard netdev stats.
v2: dropped patch "i40e: Allow disabling of DCB via debugfs" from Neerav
based on feedback from David Miller.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a couple more statistics that the hardware offers but aren't part
of the standard netdev stats.
Change-ID: I201db2898f2c284aee3d9631470bc5edd349e9a5
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The rx_errors (GLV_REPC) and rx_missed (GLV_RMPC) were removed
from the chip design.
Change-ID: Ifdeb69c90feac64ec95c36d3d32c75e3a06de3b7
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Pull the PF stat collection out of the VSI collection routine, and
add a unifying stats update routine to call the various stat collection
routines.
Change-ID: I224192455bb3a6e5dc0a426935e67dffc123e306
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This change moves some common code in two places into a
small helper function, and corrects a bug in one of the
two places in the process.
Change-ID: If3bba7152b240f13a7881eb0e8a781655fa66ce7
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Structure for VEB context added. Update macro for
transition from ms to GTIME (us) time units.
Change-ID: Ib3a19587b4cf355348655df8f60c6f37bb1497a3
Signed-off-by: Kamil Krawczyk <kamil.krawczyk@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Currently, the PF driver only notifies the VFs for PF reset events.
Instead, notify the VFs for all types of resets, so they can attempt a
graceful reinit.
Change-ID: I03eb7afde25727198ef620f8b4e78bb667a11370
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
i40evf_set_ringparam was broken in several ways. First, it only changed
the size of the first ring, and second, changing the ring size would
often result in a panic because it would change the count before
deallocating resources, causing the driver to either free nonexistent
buffers, or leak leftover buffers.
Fix this by storing the descriptor count in the adapter structure, and
updating the count for each ring each time we allocate them. This
ensures that we always free the right size ring, and always end up with
the requested count when the device is (re)opened.
Change-ID: I298396cd3d452ba8509d9f2d33a93f25868a9a55
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Hardware requires descriptors to be allocated in groups of 32.
Change-ID: I752ccc96769d1bd8d3018c004b8aeff464045bf2
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The driver was allowing the user to set larger size MTU
than the hardware was being configured to support.
The driver was already using VLAN_HLEN when setting the
hardware max receivable frame size, so just add it to the
netdev MTU set entry point as well.
Change-ID: Ie20e2a35d04f8c411253e255bea79ca69aaeaea3
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Remove unused defines and macros for RX_LRO.
Change-ID: I8ca6715edfa62b56837417a1c4ff68c2345dab6e
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Jim Young <jamesx.m.young@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
We introduced this check in case this structure changed in the future,
the AQ definition is now mature enough that this check is no longer necessary.
Change-ID: Ic66321d0a08557dc9d8cb84029185352cb534330
Signed-off-by: Kamil Krawczyk <kamil.krawczyk@intel.com>
Tested-by: Jim Young <jamesx.m.young@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Register range, being subject to register diagnostic, can vary among
different NVMs. We will try to identify the full range and use it for
a register test. This is needed to avoid false test results. If we fail
to define the proper register range we will test only the first register
from that group.
Change-ID: Ieee7173c719733b61d3733177a94dc557eb7b3fd
Signed-off-by: Kamil Krawczyk <kamil.krawczyk@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
A couple of FD checks ended up using bitwise OR to check
a value, which ends up always being evaluated to true.
This should fix the issue. Thanks to DaveJ for noticing
and reporting the issue!
CC: Dave Jones <davej@redhat.com>
Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
It is available since v3.15-rc5.
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
According to RFC1035 "[...] the total length of a domain name (i.e.,
label octets and label length octets) is restricted to 255 octets or
less."
Signed-off-by: Manuel Schölling <manuel.schoelling@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tilman Schmidt says:
====================
ISDN patches for net-next (v2)
Here's v2 of the series of patches for the ISDN CAPI subsystem
prepared by Paul Bolle and reviewed by yours truly. It reflects
GregKH's review, resulting in a substantial simplification.
Please merge via net-next.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Since v2.4 the capi driver used the following device nodes if
"middleware" support was enabled:
/dev/capi20
/dev/capi/0
/dev/capi/1
[...]
/dev/capi20 is a character device node. /dev/capi/0 (and up) are tty
device nodes (with a different major).
This device node (naming) scheme is not documented anywhere, as far as I
know. It was originally provided by the capifs pseudo filesystem (before
udev became available). It is required for example by the pppd
capiplugin. It was supported until a few years ago. But a number of
developments broke it:
- v2.6.6 (May 2004) renamed /dev/capi20 to /dev/capi and removed the
"/" from the name of capi's tty driver. The explanation of the patch
that did this included two examples of udev rules "to restore the old
namespace";
- either udev 154 (May 2010) or udev 179 (January 2012) stopped
allowing to rename device nodes, and thus the ability to have
/dev/capi20 appear instead of /dev/capi and /dev/capi/0 (and up)
instead of /dev/capi0 (and up);
- v3.0 (July 2011) also removed capifs. That disabled another method to
create the /dev/capi/0 (and up) device nodes.
So now users need to manually tweak their setup (eg, create /dev/capi/
and fill that with symlinks) to get things working. This is all rather
hacky and only discoverable by searching the web. Fix all this by
renaming /dev/capi back to /dev/capi20, and by setting the name of the
"capi_nc" tty driver to "capi!" so the tty device nodes appear as
/dev/capi/0 (and up).
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Kconfig symbol ISDN_DRV_AVMB1_VERBOSE_REASON is only used for
capi_info2str(). That function is only used in capidrv.c. So setting it
without setting ISDN_CAPI_CAPIDRV is pointless. Make it depend on
ISDN_CAPI_CAPIDRV, rename it to ISDN_CAPI_CAPIDRV_VERBOSE and put its
entry after ISDN_CAPI_CAPIDRV's entry.
Since this symbol seems to be primarily used for debugging, keep it off
by default. By now the last users of capidrv hopefully know all they
need to know about the reasons for disconnecting.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>
capi_info2str() is apparently meant to be of general utility. It is
actually only used in capidrv.c. So move it from capiutil.c to
capidrv.c and (obviously) stop exporting it.
And, since we're touching this, merge the two versions of this
function.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Herbert says:
====================
net: Support checksum in UDP
This patch series adds support for using checksums in UDP tunnels. With
this it is possible that two or more checksums may be set within the
same packet and we would like to do that efficiently.
This series also creates some new helper functions to be used by various
tunnel protocol implementations.
v2: Fixed indentation in tcp_v6_send_check arguments.
v3: Move udp_set_csum and udp6_set_csum to be not inlined
Also have this functions call with a nocheck boolean argument
instead of passing a sock structure.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Added VXLAN link configuration for sending UDP checksums, and allowing
TX and RX of UDP6 checksums.
Also, call common iptunnel_handle_offloads and added GSO support for
checksums.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call gso_make_checksum. This should have the benefit of using a
checksum that may have been previously computed for the packet.
This also adds NETIF_F_GSO_GRE_CSUM to differentiate devices that
offload GRE GSO with and without the GRE checksum offloaed.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Added a new netif feature for GSO_UDP_TUNNEL_CSUM. This indicates
that a device is capable of computing the UDP checksum in the
encapsulating header of a UDP tunnel.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call common gso_make_checksum when calculating checksum for a
TCP GSO segment.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When creating a GSO packet segment we may need to set more than
one checksum in the packet (for instance a TCP checksum and
UDP checksum for VXLAN encapsulation). To be efficient, we want
to do checksum calculation for any part of the packet at most once.
This patch adds csum_start offset to skb_gso_cb. This tracks the
starting offset for skb->csum which is initially set in skb_segment.
When a protocol needs to compute a transport checksum it calls
gso_make_checksum which computes the checksum value from the start
of transport header to csum_start and then adds in skb->csum to get
the full checksum. skb->csum and csum_start are then updated to reflect
the checksum of the resultant packet starting from the transport header.
This patch also adds a flag to skbuff, encap_hdr_csum, which is set
in *gso_segment fucntions to indicate that a tunnel protocol needs
checksum calculation
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call common functions to set checksum for UDP tunnel.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Added udp_set_csum and udp6_set_csum functions to set UDP checksums
in packets. These are for simple UDP packets such as those that might
be created in UDP tunnels.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vlad Yasevich says:
====================
Fix support for macvlan devices on top bonding
Currently, macvlan devices do not work well over bond interfaces.
Everything works well, untill a failover is triggered in the bond
device and then macvlan becomes unreachble untill arp entries
are flushed. This series adds needed functionality to
handle correct notifications and update switches with mac addresses
assigned to macvlans.
The first patch simply addes IFF_UNICAST_FLT flag to bonds since they
already correctly manage the unicast filter list of the slaves, so
we might as well prevent the bond from needlessly going into promiscuous
mode.
The second patch adds notifier handler to macvlan to trigger correct
ARP notifications.
The third patch adds handling for TLB and RLB modes that use special
ETH_P_LOOPBACK type packets to teach switch about mac addresses.
It also allow ARPs for the macvlan mac addresses to be handled by
RLB mode.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
To make TLB mode work, the patch allows learning packets
to be sent using mac addresses assigned to macvlan devices,
also taking into an account vlans that may be between the
bond and macvlan device.
To make RLB work, all we have to do is accept ARP packets
for addresses added to the bond dev->uc list. Since RLB
mode will take care to update the peers directly with
correct mac addresses, learning packets for these addresses
do not have be send to switch.
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bonding and team drivers generate specific events during failover
that trigger switch updates. When a macvlan device is configured
on top of bonding, we want switches to learn about the macvlan
devices as well. This patch adds a handler to macvlan driver to
propagate these events to all macvlan devices. We let the generic
inetdev event handler do the work.
This allows macvlan to operated correctly over active-backup
mode bond.
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bonding devices manage the unicast filters of the underlying
interfaces, but do not turn on IFF_UNICAST_FLT flag. Thus
anytime a unicast address is added to the bond, the bond is
places in promiscuous mode.
Turn on IFF_UNICAST_FLT on the bond device so that the bond does
not go into promiscuous mode needlesly. If an underlying device
does not support unicast filtering, that device will automaticall
enter promiscuous mode already.
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 30f38d2fdd.
fib_triestat is surrounded by a big lie: while it claims that it's a
seq_file (fib_triestat_seq_open, fib_triestat_seq_show), it isn't:
static const struct file_operations fib_triestat_fops = {
.owner = THIS_MODULE,
.open = fib_triestat_seq_open,
.read = seq_read,
.llseek = seq_lseek,
.release = single_release_net,
};
Yes, fib_triestat is just a regular file.
A small detail (assuming CONFIG_NET_NS=y) is that while for seq_files
you could do seq_file_net() to get the net ptr, doing so for a regular
file would be wrong and would dereference an invalid pointer.
The fib_triestat lie claimed a victim, and trying to show the file would
be bad for the kernel. This patch just reverts the issue and fixes
fib_triestat, which still needs a rewrite to either be a seq_file or
stop claiming it is.
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Antonio Ospite <ao2@ao2.it>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexander Gordeev <agordeev@redhat.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Building with CONFIG_DEBUG_SECTION_MISMATCH enabled, the following
WARNING is occured:
LD drivers/net/built-in.o
WARNING: drivers/net/built-in.o(.text+0xcd4c): Section mismatch in
reference from the function gfar_probe() to the function
.init.text:gfar_init_addr_hash_table()
The function gfar_probe() references
the function __init gfar_init_addr_hash_table().
This is often because gfar_probe lacks a __init
annotation or the annotation of gfar_init_addr_hash_table is wrong.
Signed-off-by: Xiubo Li <Li.Xiubo@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Wei Liu says:
====================
This is rebased version of Andrew's V8 patch series. The original cover letter:
--------------------
xen-net{back, front}: Multiple transmit and receive queues
This patch series implements multiple transmit and receive queues (i.e.
multiple shared rings) for the xen virtual network interfaces.
The series is split up as follows:
- Patch 1 brings the 'grant_copy_op' array back into struct xenvif, in
preparation for multi-queue support. See the patch itself for more details.
- Patches 2 and 4 factor out the queue-specific data for netback and
netfront respectively, and modify the rest of the code to use these
as appropriate.
- Patches 3 and 5 introduce new XenStore keys to negotiate and use
multiple shared rings and event channels, and code to connect these
as appropriate.
- Patch 6 documents the XenStore keys required for the new feature
in include/xen/interface/io/netif.h
All other transmit and receive processing remains unchanged, i.e. there
is a kthread per queue and a NAPI context per queue.
The performance of these patches has been analysed in detail, with
results available at:
http://wiki.xenproject.org/wiki/Xen-netback_and_xen-netfront_multi-queue_performance_testing
To summarise:
* Using multiple queues allows a VM to transmit at line rate on a 10
Gbit/s NIC, compared with a maximum aggregate throughput of 6 Gbit/s
with a single queue.
* For intra-host VM--VM traffic, eight queues provide 171% of the
throughput of a single queue; almost 12 Gbit/s instead of 6 Gbit/s.
* There is a corresponding increase in total CPU usage, i.e. this is a
scaling out over available resources, not an efficiency improvement.
* Results depend on the availability of sufficient CPUs, as well as the
distribution of interrupts and the distribution of TCP streams across
the queues.
Queue selection is currently achieved via an L4 hash on the packet (i.e.
TCP src/dst port, IP src/dst address) and is not negotiated between the
frontend and backend, since only one option exists. Future patches to
support other frontends (particularly Windows) will need to add some
capability to negotiate not only the hash algorithm selection, but also
allow the frontend to specify some parameters to this.
Note that queue selection is a decision by the transmitting system about
which queue to use for a particular packet. In general, the algorithm
may differ between the frontend and the backend with no adverse effects.
Queue-specific XenStore entries for ring references and event channels
are stored hierarchically, i.e. under .../queue-N/... where N varies
from 0 to one less than the requested number of queues (inclusive). If
only one queue is requested, it falls back to the flat structure where
the ring references and event channels are written at the same level as
other vif information.
V8:
- Squash the queue error handling code into patch 3.
- Update the documentation (patch 6) according to comments on the
equivalent patch to Xen.
V7:
- Rebase on latest net-next, which includes the netback grant mapping
patch series from Zoltan Kiss
- Reduce QUEUE_NAME_SIZE by 1 to avoid double-counting the trailing '\0'
- Simplify the queue hashing by using (hash % num_queues) instead of
multiply & shift.
- Add ratelimited warning for invalid queue selection.
- Fix error handling to correctly tear down already setup queues.
- Use dev->real_num_tx_queues instead of separately maintaining a
count of the number of queues.
V6:
- Use 'max_queues' as the module param. name for both netback and netfront.
V5:
- Fix bug in xenvif_free() that could lead to an attempt to transmit an
skb after the queue structures had been freed.
- Improve the XenStore protocol documentation in netif.h.
- Fix IRQ_NAME_SIZE double-accounting for null terminator.
- Move rx_gso_checksum_fixup stat into struct xenvif_stats (per-queue).
- Don't initialise a local variable that is set in both branches (xspath).
V4:
- Add MODULE_PARM_DESC() for the multi-queue parameters for netback
and netfront modules.
- Move del_timer_sync() in netfront to after unregister_netdev, which
restores the order in which these functions were called before applying
these patches.
V3:
- Further indentation and style fixups.
V2:
- Rebase onto net-next.
- Change queue->number to queue->id.
- Add atomic operations around the small number of stats variables that
are not queue-specific or per-cpu.
- Fixup formatting and style issues.
- XenStore protocol changes documented in netif.h.
- Default max. number of queues to num_online_cpus().
- Check requested number of queues does not exceed maximum.
--------------------
I rebased this on top of net-next. No functional change is introduced. The
patch that needed some extra care was "xen-netback: Factor queue-specific data
into queue struct" because it clashed with a fix introduced in net. A simple
test of creating guest, iperf, then shutting down guest worked as expected.
The last patch fixes a minor problem that queue name is not initialised in
xen-netfront, resulting in names like "-tx" "-rx" in /proc/interrupt.
Changes since v9 (no functional change introduced):
* include commit summary in the commit message of first patch
* fold David Vrabel's Reviewed-by into last patch
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
