| Age | Commit message (Collapse) | Author |
|---|
|
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Also try to prevent some possible slave dereference race conditions. This is
finalized in the next patch, which abandons the slave array in favour of
a list_head list and list RCU.
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Also move the frame receive handler to hsr_slave.c.
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When the system is too busy to complete the urb, the tx timout function
would be called. This causes the other tx urbs would be killed, too.
Increase the tx timeout to avoid it.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
ic_dev_xid is only used in ipconfig.c
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: netdev@vger.kernel.org
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Tom Lendacky says:
====================
amd-xgbe: AMD 10Gb Ethernet driver updates
The following series fixes some bugs and provides new/changed support
in the driver.
- Fix a debugfs backward compatibility issue introduced by a previous patch
- Write to the interrupt enablement register, not the status register when
setting MTL interrupts
- Call netif_napi_del whenever the ndo_stop operation is called (to match
the call to netif_napi_add on ndo_open)
- Peformance enhancements:
- Adjusted default coalescing settings
- AXI DMA changes (burst length size and cache settings)
- ioread/iowrite reduction during interrupt
- Napi poll updates
- AXI DMA settings based on device tree property to account for a change in
the ARM64 default cache operations assignment
This patch series is based on net-next.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The default cache operations for ARM64 were changed during 3.15.
To use coherent operations a "dma-coherent" device tree property
is required. If that property is not present in the device tree
node then the non-coherent operations are assigned for the device.
Add support to the amd-xgbe driver to assign the AXI DMA cache settings
based on whether the "dma-coherent" property is present in the device
node. If present, use settings that work with the caches. If not
present, use settings that do not look at the caches.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch provides some general performance enhancements for the
driver:
- Modify the default coalescing settings (reduce usec, increase frames)
- Change the AXI burst length to 256 bytes (default was 16 bytes which
was smaller than a cache line)
- Change the AXI cache settings to write-back/write-allocate which
allocate cache entries for received packets during the DMA since the
packet will be processed soon afterwards
- Combine ioread/iowrite when disabling both the Tx and Rx interrupts
- Change to processing the Tx/Rx channels in pairs
- Only recycle the Rx descriptors when a threshold of dirty descriptors
is reached
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Currently the napi context is added using netif_napi_add each time
the ndo_open operation is called. However, there is not a
corresponding netif_napi_del call during the ndo_stop operation. If
the device ndo_open operation was called more than once an infinite
loop occurs during module unload. Add a call to netif_napi_del during
the ndo_stop operation.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When initializing the MTL interrupts the interrupt status
register is written to instead of the interrupt enable register.
Since no MTL interrupts are being enabled and the default state
is for MTL interrupts to be disabled this did not cause a problem,
but needs to be fixed to target the correct register.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The initial change from sscanf to kstrtouint broke backward
compatbility by using a base of "0" in the kstrtouint call.
This allowed for entering decimal, hexadecimal or octal as
input where previously the sscanf always interpreted the input
as hexadecimal. Additionally, -EIO was returned on error prior
to this change and now it is whatever the error value that is
returned by kstrtouint.
Change the base value of the kstrtouint from 0 to 16 and return
-EIO on error.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Reported-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The header file include/linux/arcdevice.h #defines bool to int, if
bool is not already #defined. However, the files which use that header
file seem to rely on that #define (unconditionally) being in effect:
the prototypes for the functions arcrimi_reset, com20020_reset,
com90io_reset, com90xx_reset (whose addresses are assigned to the
hw.reset member of struct arcnet_local) use int explicitly.
Moreover, that #define is an accident waiting to happen (scenario:
inclusion of arcdevice.h followed by inclusion of some header which
declares function prototypes using bool). Also, #include
<linux/types.h> must appear before #include <linux/arcdevice.h> (the
compiler wouldn't like "typedef _Bool int").
Since none of the files using arcdevice.h declare variables of type
"bool", the patch is actually quite simple, unlike the commit message.
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
enic_set_coalesce() has two problems.
* It should return -EINVAL and not -EOPNOTSUPP for invalid coalesce values.
* In case of MSIX, enic_set_coalesce return error after applying requested
coalescing setting partially. We should either apply all the setting requeste
and return success or apply non and return error.
* This patch also simplifies the algo.
This was introduced by
'7c2ce6e60f703 enic: Add support for adaptive interrupt coalescing'
These changes were suggested by Ben Hutchings here
http://www.spinics.net/lists/netdev/msg283972.html
Also change enic driver version.
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
These warnings are no longer relevant. Even when last slave is
removed, there is a valid address assigned to bond (random).
The correct functionality of vlans is ensured by maintaining unicast
list in vlan_sync_address().
Suggested-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Veaceslav Falico <vfalico@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Alexander Aring says:
====================
at86rf230: rework driver implementation
this patch series includes a rework of the at86rf230 driver.
There are several changes:
- Add regmap support.
- Merge at86rf212 operations with generic at86rf2xx operations, all chips
supports these operations.
- Drop of irqworker. This is a workqueue which will scheduled by an irq to
handle synchronous spi handling. Instead using asynchronous spi handling,
then no scheduler is involved at irq handling.
- Also detected some bugs by receiving frame like CRC can be correct and a
802.15.4 frame length could be above 127 bytes. This would crash the whole
kernel (but should be handled by the mac layer). Another bug is the handling
with RX_SAFE_MODE which protect the frame buffer after a readout. This is
currently not working because we read out the buffer twice and the first one
to get the frame size. Solution is to readout always the whole frame buffer.
- Added some timing relevants things from the datasheet for state changes And
IEEE 802.15.4 standard like interframe spacing. Interframe spacing is needed
to insert some receiving space time between frame transmitting. This should be
also handled by MAC layer, but it's currently a workaround to add this inside
the driver layer.
- Add some callback setting for chip specific handling, instead of runtime decisions
if (is_chip_type()). Callbacks are set only once at probe time.
- We don't using a force state change anymore. A force state change will do a
abort of receiving frames while we want to transmit a new frame. This should
decrease the drop rate of packets.
- And many others changes and bug fixes...
changes since v3:
- fix irq polarity in patch ("at86rf230: rework irq_pol setting").
changes since v2:
- add check if necessary functions are implemented when hw flags are set in patch
("mac802154: at86rf230: add hw flags and merge ops"). I choosed the second variant.
- remove unnecessary includes for workqueue and mutex in patch
("at86rf230: rework transmit and receive").
- remove unnecessary cast in patch ("at86rf230: rework transmit and receive").
- acivate regmap cache with REGCACHE_RBTREE in patch
("at86rf230: add regmap support").
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch removes the current synchron state change function and add a
new function for a state assert. Change the start and stop callbacks to
use this new synchron state change behaviour. It's a wrapper around the
async state change function.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch rework the irq_pol register setting for rising and falling
interrupt settings only. The default behaviour should be rising flag.
Also use IRQ_TYPE_* defines instead of IRQF_* defines. There is no
functionality change but irq_get_trigger_type returns IRQ_TYPE_* defines.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
There is no need to set this bit in start callback which could be
called more than once.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch is a complete reimplementation of transmit and receive
handling for the at86rf230 driver.
It solves also six bugs:
First:
The RX_SAFE_MODE is enabled and the transceiver doesn't leave the
receive state while the framebuffer isn't read by a CMD_FB command.
This is useful to read out the frame and don't get into another receive
or transmit state, otherwise the frame would be overwritten.
The current driver do twice CMD_FB calls, the first one leaves this
protection.
Second:
Sometimes the CRC calculation is correct and the length field is greater
127. The current mac802154 layer and filter of a at86rf2xx doesn't check
on this and the kernel crashes. In this case the frame is corrupted, we
send the whole receive buffer to the next layer which can be useful for
sniffing.
Thrid:
There is a undocumented race condition. When we are go into the
RX_AACK_ON state the transceiver could be changed into RX_AACK_BUSY
state. This is a normal behaviour. In this case the transceiver received
a SHR while assert wasn't finished.
Fourth:
It also handle some more "correct" state changes. In aret mode the
transceiver need to go to TX_ON before the transceiver go into
RX_AACK_ON.
Fifth:
The programming model [0] describes also a error handling in ARET mode
if the trac status is different than zero. This is patch adds support
for handling this.
Sixth:
In receive handling the transceiver should also get the trac status
according [0]. The driver could use the trac status as error statistic
handling, but the driver doesn't use this currently. There is maybe some
timing behaviour or the read of this register change some transceiver
states.
In addition the irqworker is removed. Instead we do async spi calls and
no scheduling is involved anymore. The transmit function is also
asynchron but with a wait_for_completion handling. The mac802154 layer
doesn't support asynchron transmit handling right now.
The state change behaviour is now changes, before it was:
1. assert while(!STATE_TRANSITION_IN_PROGRESS)
2. state change
3. assert while(!STATE_TRANSITION_IN_PROGRESS)
4. assert once(wanted state != current state)
Sometimes a unexcepted state change occurs when 4. assert was violated.
The new state change behaviour is:
1. assert while(!STATE_TRANSITION_IN_PROGRESS)
2. state change
3. wait state change timing according datasheet
4. assert once(wanted state != current state)
This behaviour is described in the at86rf231 software programming model [0].
The state change documentation in this programming guide should also valid for
at86rf212 and at86rf233 chips.
The transceiver don't do a FORCE_TX_ON while we want to transmit a PDU.
The new behaviour is a TX_ON and wait a receiving time (tFrame + tPAck).
If we are still in RX_AACK_BUSY then we transmit a FORCE_TX_ON as timeout
handling. The different is that FORCE_TX_ON aborts receiving and TX_ON
waits if RX_AACK_BUSY is finished. This should decrease the drop rate of
packets.
[0] http://www.atmel.com/Images/AVR2022_swpm231-2.0.zip
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
To set the CCA_ED_THRES register the calculation for at86rf23x is
different than for at86rf212. This patch adds a new callback for this
calculation in chip data struct.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds a new at86rf2xx_chip_data structure which holds device
specific attributes. Instead of runtime decisions "if (is212())" we set
callbacks/attributes while device detection.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch drops the current lowlevel spi calls for the detect device
function instead we handle this via regmap. Also put the detection of
in a seperate function and set all device specific attributes while detection.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds regmap support for the at86rf230 driver and drop the
lowlevel spi access functions and use the regmap access functions.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds new mac802154 hw flags for transmit power, csma and
listen before transmit (lbt). These flags indicates that the transceiver
supports these features. If the flags are set and the driver doesn't
implement the necessary functions, then ieee802154_register_device
returns -ENOSYS "Function not implemented".
This patch merges also all at86rf230 operations into one operations structure
and set the right hw flags for the at86rf230 transceivers.
Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
Jeff Kirsher says:
====================
Intel Wired LAN Driver Updates 2014-07-02
This series contains updates to i40e and i40evf.
Anjali fixes a possible race where we were trying to free the dummy packet
buffer in the function that created it, so cleanup the dummy packet buffer
in i40e_clean_tx_ring() instead. Also fixes an issue where the filter
program routine was not checking if there were descriptors available for
programming a filter.
Mitch fixes unnecessary delays when sending the admin queue commands by
moving a declaration up one level so we do not dereference it out of scope.
Fixes an issue with the VF where if the admin queue interrupts get lost for
some reason, the VF communication will stall as the VFs have no way of
reaching the PF. To alleviate this condition, go ahead and check the ARQ
every time we run the service task. Updates i40evf to allow the watchdog
to fire vector 0 via software, which makes the driver tolerant of dropped
interrupts on that vector.
Paul fixes a shifted '1' to be unsigned to avoid shifting a signed integer.
Jesse disables TPH by default since it is currently not enabled in the
current hardware. Also finishes the i40e implementation of get_settings
for ethtool.
Catherine adds a new variable (hw.phy.link_info.an_enabled) to track whether
auto-negotiation is enabled, along with the functionality to update the
variable. Adds the functionality to set the requested flow control mode.
Adds i40e implementation of setpauseparam and set_settings to ethtool.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Russell King says:
====================
Freescale ethernet driver updates
Here's the first batch of patches for the Freescale FEC ethernet driver.
They require the previously applied "net: fec: Don't clear IPV6 header
checksum field when IP accelerator enable" patch.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
fec_enet_alloc_buffers() assumes that kmalloc() will never fail, which
is an invalid assumption. Fix this by implementing a common error
cleanup path, and use it to also clean up after failed bounce buffer
allocation.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Ensure that we do not double-free any allocations, and that any transmit
skbuffs are properly freed when we clean up the rings.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Avoid writing any state until we're certain we can proceed with the
transmission: this avoids writing mapping error address values to the
descriptors, or setting the skbuff pointer until we have successfully
mapped the skb.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Allocate, and then map the receive skb before writing any data to the
ring descriptor or storing the skb. When freeing the receive ring
entries, unmap and free the skb, and then clear the stored skb pointer.
This means we have ring data and skb pointer in one of two states:
either both fully setup, or nothing setup.
This simplifies the cleanup, as we can use just the skb pointer to
indicate whether the descriptor is setup, and thus avoids potentially
calling dma_unmap_single() on a DMA error value.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
napi_disable() waits until the NAPI processing has completed, and then
prevents any further polls. At this point, the driver then clears
fep->opened. The NAPI poll function uses this to stop processing in
the receive path. Hence, it will never see this variable cleared,
because the NAPI poll has to complete before it will be cleared.
Therefore, this variable serves no purpose, so let's remove it.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When the network interface goes down, stop the phy to prevent further
link up status changes before taking the MAC or netif sections down.
This prevents further reception of link up events which could
potentially call fec_restart().
Since phy_stop() takes the mutex which adjust_link() runs under, we
also ensure that adjust_link() will not already be processing a link
up event.
We also need to do this when suspending as well - we don't want a
mis-timed phy state change to restart the MAC after we have stopped
it for suspend, and thus need to restart the phy when resuming.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When we disconnect from a phy, we should forget our pointer to it so we
don't accidentally try to configure it. We handle a NULL phy pointer
correctly in most places, except fec_enet_set_pauseparam(). Fix this
too.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
fep->phy_dev can not be NULL here for two reasons:
- fec_enet_open() will have successfully connected the phy, or will have
failed.
- fec_enet_open() will have called phy_start(fep->phy_dev), which
unconditionally dereferences this pointer.
If it were to be NULL here, then fec_enet_open() will have already
oopsed.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We use netif_stop_queue() in several places where we want to ensure that
the start_xmit function is not running. netif_stop_queue() is not
sufficient to achieve that - it merely sets a flag to indicate that the
transmit queue(s) should not be run.
netif_tx_disable() gives this guarantee, since it takes the transmit
queue lock while marking the queue stopped. This will wait for the
transmit function to complete before returning.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
While running: while :; do iperf -c <HOST> -P 4; done, transmit timeouts
are regularly reported. With the tx ring dumping in place, we can see
that all entries are in use, and the hardware has finished transmitting
these packets. However, the driver has not reclaimed these ring
entries.
This can occur if the interrupt handler is invoked at the wrong moment -
eg:
CPU0 CPU1
fec_enet_tx()
interrupt, IEVENT = FEC_ENET_TXF
FEC_ENET_TXF cleared
napi_schedule_prep()
napi_complete()
The result is that we clear the transmit interrupt, but we don't trigger
any cleaning of the transmit ring. Instead, use a different strategy:
- When receiving a transmit or receive interrupt, disable both tx and rx
interrupts, but do not acknowledge them. Schedule a napi poll. Don't
loop.
- When we are polled, read IEVENT, acknowledging the pending transmit
and receive interrupts, before then going on to process the
appropriate rings.
This allows us to avoid the race, and has a number of other advantages:
- we cut down on the number of transmit interrupts we have to process.
- we only look at the rings which have pending events.
- we gain additional throughput: the iperf total bandwidth increases
from about 180Mbps to 240Mbps:
[ 3] 0.0-10.0 sec 68.1 MBytes 57.0 Mbits/sec
[ 5] 0.0-10.0 sec 72.4 MBytes 60.5 Mbits/sec
[ 4] 0.0-10.1 sec 76.1 MBytes 63.5 Mbits/sec
[ 6] 0.0-10.1 sec 71.9 MBytes 59.9 Mbits/sec
[SUM] 0.0-10.1 sec 288 MBytes 241 Mbits/sec
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Setting the pause parameters causes a running network interface to be
restarted. However, the restart forces the FEC into half-duplex mode,
whether or not the remote end is in half-duplex mode. Misconfigured
duplex mode is a known source of problems on a link.
Fix this by always preserving the duplex mode on configuration changes.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The iMX6 gigabit FEC does not support half-duplex gigabit operation.
Phys attacked to the FEC may support this, and we currently do nothing
to disable this feature. This may result in an invalid configuration.
Mask out phy support for gigabit half-duplex operation.
Acked-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Tom Herbert says:
====================
net: Improvements and applications of packet flow hash in transmit path
This patch series includes some patches which improve and make use
of skb->hash in the transmit path.
What is included:
- Infrastructure to save a precomputed hash in the sock structure.
For connected TCP and UDP sockets we only need to compute the
flow hash once and not once for every packet.
- Call skb_get_hash in get_xps_queue and __skb_tx_hash. This eliminates
the awkward access to skb->sk->sk_hash in the lower transmit path.
- Move UDP source port generation into a common function in udp.h This
implementation is mostly based on vxlan_src_port.
- Use non-zero IPv6 flow labels in flow_dissector as port information
for flow hash calculation.
- Implement automatic flow label generation on transmit (per RFC 6438).
- Don't repeatedly try to compute an L4 hash in skb_get_hash if we've
already tried to find one in software stack calculation.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add sw_hash flag to skbuff to indicate that skb->hash was computed
from flow_dissector. This flag is checked in skb_get_hash to avoid
repeatedly trying to compute the hash (ie. in the case that no L4 hash
can be computed).
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Automatically generate flow labels for IPv6 packets on transmit.
The flow label is computed based on skb_get_hash. The flow label will
only automatically be set when it is zero otherwise (i.e. flow label
manager hasn't set one). This supports the transmit side functionality
of RFC 6438.
Added an IPv6 sysctl auto_flowlabels to enable/disable this behavior
system wide, and added IPV6_AUTOFLOWLABEL socket option to enable this
functionality per socket.
By default, auto flowlabels are disabled to avoid possible conflicts
with flow label manager, however if this feature proves useful we
may want to enable it by default.
It should also be noted that FreeBSD has already implemented automatic
flow labels (including the sysctl and socket option). In FreeBSD,
automatic flow labels default to enabled.
Performance impact:
Running super_netperf with 200 flows for TCP_RR and UDP_RR for
IPv6. Note that in UDP case, __skb_get_hash will be called for
every packet with explains slight regression. In the TCP case
the hash is saved in the socket so there is no regression.
Automatic flow labels disabled:
TCP_RR:
86.53% CPU utilization
127/195/322 90/95/99% latencies
1.40498e+06 tps
UDP_RR:
90.70% CPU utilization
118/168/243 90/95/99% latencies
1.50309e+06 tps
Automatic flow labels enabled:
TCP_RR:
85.90% CPU utilization
128/199/337 90/95/99% latencies
1.40051e+06
UDP_RR
92.61% CPU utilization
115/164/236 90/95/99% latencies
1.4687e+06
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch implements the receive side to support RFC 6438 which is to
use the flow label as an ECMP hash. If an IPv6 flow label is set
in a packet we can use this as input for computing an L4-hash. There
should be no need to parse any transport headers in this case.
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
