gro: Optimise TCP packet reception
As this function can be called more than half a million times for
10GbE, it's important to optimise it as much as we can.
This patch uses bit ops to logical ops, as well as open coding
memcmp to exploit alignment properties.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
As this function can be called more than half a million times for
10GbE, it's important to optimise it as much as we can.
This patch does some obvious changes to use 2-byte and 4-byte
operations instead of byte-oriented ones where possible. Bit
ops are also used to replace logical ops to reduce branching.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch optimises the Ethernet header comparison to use 2-byte
and 4-byte xors instead of memcmp. In order to facilitate this,
the actual comparison is now carried out by the callers of the
shared dev_gro_receive function.
This has a significant impact when receiving 1500B packets through
10GbE.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch prepares for the move of the same_flow checks out of
dev_gro_receive. As such we need to remember the number of held
packets since doing a loop just to count them every time is silly.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Several devices need to insert some "pre headers" in front of the
main packet data when they transmit a packet.
Currently we allocate only 16 bytes of pad room and this ends up not
being enough for some types of hardware (NIU, usb-net, s390 qeth,
etc.)
So increase this to 32.
Note that drivers still need to check in their transmit routine
whether enough headroom exists, and if not use skb_realloc_headroom().
Tunneling, IPSEC, and other encapsulation methods can cause the
padding area to be used up.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update the version number to 1.3.16 and update copyright dates for 2009.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch addresses two minor items I found while cleaning up the igb
driver for our sourceforge version.
The first clears the context index if we don't flag that we need it.
The second item is that eims_other should be used instead of bit defines
when setting all of the EICS bits prior to reset.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It was seen with repeated interface up/down testing that there was a large
stray between the stats reported by the queues and the stats reported by the
HW. It was found to be an issue in that hw stats were being reset without
first being recorded. This change records the stats before wiping them from
the system via the reset.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the setting of ring->count variables from igb_probe as they are
duplicating the same configuration that is done igb_alloc_queues.
Remove the err_hw_init tag as it can be replaced by err_sw_init.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The disable_av variable is never used by the driver and provides no value as
it is likely a leftover debugging variable. I have removed it and replaced
the one spot that checked for it with a check for a valid address.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As additional hardware is added to the igb driver it is easier to support
the expansion via switch statements instead of using nested ifs. For
this reason I am changing this to a switch statement.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move igb_get_hw_control up so that it is called just after the reset in
igb_resume. This notifies the HW sooner that the driver is reassuming
control of the device.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The interrupt handler was reading eicr and then doing nothing with the
result. I have removed the variable and the register read since they
provide no value to the legacy interrupt handler.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There were several spots in the code making calls to adapter->hw when they
could have just been accessing hw-> directly. I cleaned up the spots where
this was visibly apparent.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There isn't much point in having the _82575 hanging off the end of this
function since there aren't any other version of this function running
around within this driver. This also allows for a bit of whitespace
cleanup due to a shorter function name.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The igb watchdog task is modifying the watchdog timer twice duing a single
run. It only needs to be called once to reschedule itself for 2 seconds from
the last time it ran.
In addition I removed the allocation of the mac_info structure since it is
only called twice and is easier to access via the e1000_hw struct.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch cleans up igb_netpoll so that it is more friendly with both the
current napi and newly introduced GRO features.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a counter for dma out of sync errors reported via interrupt.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most of the code for the testing has pretty much become stale at this point
and is need of update. This update just streamlines most of the code,
widens the range of interrupt testing.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This driver is currently using HW_CSUM which is not correct. Update this
to use the IP_CSUM and IPV6_CSUM flags. In addition consolidate the TSO
flag setting.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This statistic is not used and so it is safe to remove
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
All of the nvm ops have the tag _nvm added to the end which is redundant
since all of the calls to the ops have to go through the nvm ops struct
anyway.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch renames write_phy_reg to write_reg and read_phy_reg to read_reg.
It seems redundant to call out phy in an operation that is part of the
phy_ops struct.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of pulling the mac address from EEPROM it is easier to pull it from
the RAL/RAH registers and then just copy it into the address structures.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes dev_spec a union and simplifies it so that it does not
require dynamic allocation and freeing in the driver.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a link check function to contain all activities related to verifying
that the link is present. The current approach is a bit cludgy and needs
to be cleaned up.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since igb supports only pci-e nics and there is no plan to support any
legacy pci parts in the driver there isn't really much need for checking to
see if an io port is needed.
In the unlikely event that we do begin supporting legacy pci parts then we
can see about adding this code back to the driver.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch moves the initialization of the number of queues into
set_interrupt_capability. This allows the number of queues to increase in
the unlikely event that the system initially fails to allocate enough msi-x
interrupts, does a suspend/resume, and then can allocate enough interrupts
on resume.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
buffsz is being repeatedly set when allocaing buffers. Since this value
should only need to be set once in the function I am moving it out of the
looped portion of the path.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While cleaning up the skb_over panic with small frames I found there was
room for improvement in the ordering of operations within the rx receive
flow. These changes will place the prefetch for the next descriptor to a
point earlier in the rx path.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unfortunately, the OF device tree nodes for SBUS and PCI
hme devices have the same device node name on some systems.
So if the name of the parent node isn't 'sbus', skip it.
Based upon an excellent report and detective work by
Meelis Roos and Eric Brower.
Signed-off-by: David S. Miller <davem@davemloft.net>
Tested-by: Meelis Roos <mroos@linux.ee>
Now that our set_num_queues() routines for each feature are re-entrant, and
can be called at any point, they shouldn't zero out the feature's indices
or mask bits. Subsequent calls into those routines for those features can
result in zero Rx and Tx queues being assigned, causing a panic later in
driver reinitialization.
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds another type of recoverable error to the driver. It also
modifies the sequence for recovery to include a mac reset and clearing
of interrupts.
Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the ethtool tx csum "set" command. A recent patch was
submitted to remove HW_CSUM and use IP_CSUM instead. Therefore, the
corresponding ethtool command should also be modified.
Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes an issue with the suspend/resume cycle with msi
interrupts. See bugzilla number 10487 for more details. The fix is to
re-setup a private msi pci config offset field.
Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch updates the logic used to communicate with the mgmt unit. It
also adds a version check for a newer mgmt unit firmware.
* Fixed udelay to schedule_timeout_uninterruptible
Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the changes were done to the protocol last release, some endian
bugs crept in. This patch fixes those endian problems and has been
verified to run on 32/64 bit and x86/ppc architectures.
This version of the patch incorporates the correct annotations
for endian variables.
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based upon a patch from Atsushi Nemoto <anemo@mba.ocn.ne.jp>
--------------------
The commit 649274d993 ("net_dma:
acquire/release dma channels on ifup/ifdown") added unconditional call
of dmaengine_get() to net_dma. The API should be called only if
NET_DMA was enabled.
--------------------
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Dan Williams <dan.j.williams@intel.com>
From: Ondrej Zary <linux@rainbow-software.org>
last year, I posted a patch which fixed hibernation on 3c509
cards. That was back in 2.6.24. It worked fine in 2.6.25. But then I
stopped using hibernation (as it did not work with my new IT8212 RAID
controller).
Now I fixed it and noticed that 3c509 does not wake up properly
anymore (in 2.6.28) - neither in PnP nor in ISA modes. ifconfig
down/up makes the card work again in PnP mode. However, in ISA mode,
ifconfig up ends with "No such device" error.
Comparing the 3c509 driver between 2.6.25 and 2.6.28, there's only
some statistics-related change. So the cause of the problem must be
somewhere else.
This patch makes the resume work in PnP mode, but it's still not
enough for ISA mode.
Signed-off-by: David S. Miller <davem@davemloft.net>
From: Ilkka Virta <itvirta@iki.fi>
In the lockup situation the driver seems to go off in an eternal storm
of interrupts right after calling request_irq(). It doesn't actually
do anything interesting in the interrupt handler. Since connecting the link
afterwards works, something later in initialization must fix this.
Looking at gem_do_start() and gem_open(), it seems that the only thing
done while opening the device after the request_irq(), is a call to
napi_enable().
I don't know what the ordering requirements are for the
initialization, but I boldly tried to move the napi_enable() call
inside gem_do_start() before the link state is checked and interrupts
subsequently enabled, and it seems to work for me. Doesn't even break
anything too obvious...
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a potential NULL dereference bug during error handling in
rxrpc_kernel_begin_call(), whereby rxrpc_put_transport() may be handed a NULL
pointer.
This was found with a code checker (http://repo.or.cz/w/smatch.git/).
Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some Realtek chips (RTL8169sb/8110sb in my case) are unable to retrieve
ethtool statistics when the interface is down. The process stays in
endless loop in rtl8169_get_ethtool_stats. This is because these chips
need to have receiver enabled (CmdRxEnb bit in ChipCmd register) that is
cleared when the interface is going down. It's better to update statistics
only when the interface is up and otherwise return copy of statistics
grabbed when the interface was up (in rtl8169_close).
It is interesting that PCI-E NICs (like 8168b/8111b...) are not affected.
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Herbert Xu