DCA related fixes
=================
- ixgbe was not compiling and using DCA correctly if dca was a module
- DCA interface changed with new kernel
- ixgbe was not correctly configured to indicate DCA hints to the
correct CPU.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
as mentioned by Herbert, our hardware supports IP offloads, not full
checksum offloads for any protocol in existence (even though the
hardware just provides generic csum support over any range of bytes)
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
This locks the seed down so loading/unloading the driver will present
predictable hashing from RSS. Also move the rx_buf_len out of the adapter
struct, and into the Rx ring struct.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
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: Jeff Garzik <jgarzik@redhat.com>
Enable Tx Head Writeback in the hardware. This helps performance by
removing adapter writebacks to descriptors on transmit completion.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
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: Jeff Garzik <jgarzik@redhat.com>
Since the adapter cannot tell what the remote end's flow control capability
is through auto-neg, we must turn off flow control by default.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Cleanup all the different references to the Tx ring and Rx ring structures
and make them common across the driver.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
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: Jeff Garzik <jgarzik@redhat.com>
Implement HAVE_SET_RX_MODE in the driver for MC and UC lists.
Signed-off-by: Christopher Leech <christopher.leech@intel.com>
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: Jeff Garzik <jgarzik@redhat.com>
the driver was misinterpreting rx_csum return value in the descriptor
so occassionally we would indicate an rx_csum error in our stats
when there was none. This would have no effect on traffic because
we would just hand the packet to the stack anyway without the offload
flag set, but would increase CPU for those packets that needed a recompute.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
VLAN filtering is broken, due to reading the incorrect register for
the VLAN filtering settings. Fixed by reading/writing the correct
register.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
This is a simple device ID add for adapters that support
CX4 (copper infiniband style cable) connectors for 10GbE.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
The ixgbe driver was untested with device ID 8086:10c8 but still advertises
support. Currently if this device is present in the system when the driver
is loaded, the system will panic.
Remove this device ID until full support can be tested with available
hardware. This patch is necessary for 2.6.24, 2.6.25 and 2.6.26
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
With the recent changes to tx mutiqueue, igb/ixgbe/e1000e was not calling
netif_tx_start_all_queues() before calling netif_tx_wake_all_queues().
This causes an issue during loading of the driver.
In addition, updated e1000e to use the updated tx mutliqueue api.
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This effectively "flips the switch" by making the core networking
and multiqueue-aware drivers use the new TX multiqueue structures.
Non-multiqueue drivers need no changes. The interfaces they use such
as netif_stop_queue() degenerate into an operation on TX queue zero.
So everything "just works" for them.
Code that really wants to do "X" to all TX queues now invokes a
routine that does so, such as netif_tx_wake_all_queues(),
netif_tx_stop_all_queues(), etc.
pktgen and netpoll required a little bit more surgery than the others.
In particular the pktgen changes, whilst functional, could be largely
improved. The initial check in pktgen_xmit() will sometimes check the
wrong queue, which is mostly harmless. The thing to do is probably to
invoke fill_packet() earlier.
The bulk of the netpoll changes is to make the code operate solely on
the TX queue indicated by by the SKB queue mapping.
Setting of the SKB queue mapping is entirely confined inside of
net/core/dev.c:dev_pick_tx(). If we end up needing any kind of
special semantics (drops, for example) it will be implemented here.
Finally, we now have a "real_num_tx_queues" which is where the driver
indicates how many TX queues are actually active.
With IGB changes from Jeff Kirsher.
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no need for a feature bit for something that
can be tested by simply checking the TX queue count.
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently VLAN filtering is enabled when the first VLAN is added.
Obviously before that there's no point in receiving any VLAN packets.
Now that we disable VLAN filtering in promiscous mode, we can keep
the VLAN filters enabled the remaining time.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As discussed in this thread:
http://www.mail-archive.com/netdev@vger.kernel.org/msg53976.html
promiscous mode means to disable *all* filters. Currently only unicast
and multicast filtering is disabled. This patch changes all Intel
drivers to also disable VLAN filtering.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
EEh is not recovering in a resonable amount of time on PPC during
ixgbe_down().
Signed-off-by: Paul Larson <pl@us.ibm.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Support for in-kernel LRO with the ability to enable/disable via ethtool
based on comments from Ben Hutchings.
Signed-off-by: Mallikarjuna R Chilakala <mallikarjuna.chilakala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: PJ Waskiewicz <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Using the new interface for propagating device feature flags into VLAN
devices, turn on TSO and CSUM offload on VLAN devices.
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: Jeff Garzik <jgarzik@redhat.com>
To enable EEH support for pci-express network adapters, pcie/msi state
needs to be saved and restored for that adapter.
Tested this EEH patch with Intel 10G pci-express ixgbe adapter.
Signed-off-by: Wendy Xiong <wendyx@us.ibm.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Under CONFIG_DCA the compilation is broken since the commit
bd0362dde0 (ixgbe: Add optional
DCA infrastructure).
IXGBE_SUCCESS is not defined anywhere, replace it with 0.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
82598 cards and up support DCA, which enables the chipset to warm
up the caches for upcoming payload data. This code makes the
driver plug into the CONFIG_DCA infrastructure that was merged
earlier.
Signed-off-by: Jeb Cramer <cramerj@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
82598 can produce a formidable interrupt rate, and is largely
unusable without some form of moderation. The default behaviour
before this patch is to limit irq's to a reasonable number.
However, just like our other drivers we can reduce latency
for small packet-type traffic considerably by allowing the
irq rate to go up dynamically.
This patch introduces a simple irq moderation algorithm based
on traffic analysis. The driver will use more CPU to service
small packets quicker but will perform the same on bulk traffic
as the old code.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Now that the irq vector code is in place, we can add the conditional
multiqueue TX code in the driver. This requires the optional
CONFIG_NETDEVICES_MULTIQUEUE=y and will not be enabled without
it.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Acked-by: Waskiewicz Jr, Peter P <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
This code abstracts the per-queue MSI-X interrupt vector into
a queue vector layer. This abstraction is needed since there can
be many more queues than available MSI-X vectors in a machine.
The MSI-X irq vectors are remapped to a shared queue vector which
can point to several (both RX and TX) hardware queues. The NAPI
algorithm then cleans the appropriate ring/queues on interrupt
or poll.
The remapping is a delicate and complex calculation to make sure
that we're not unbalancing the irq load, and spreads the irqs
as much as possible, and may combine RX and TX flows onto the
same queue vector.
This effectively enables receive flow hashing across vectors
and helps irq load balance across CPUs.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Acked-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: Waskiewicz Jr, Peter P <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
In reply to "RE: [Fwd: [PATCH 2.6.25] ixgbe/igb: correctly obtain protocol
information on transmit]" from Andy Gospodarek:
The driver was incorrectly looking at socket headers for
protocol information, needed for checksumming offload. Fix
this by not looking at the socket but frame headers instead.
This disregards extension headers but it's unclear that linux
generates those anyway.
Tested by Andy Gospodarek.
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
It's easy to oversee this issue when working with this card
as evrything will work OK but performance is severely limited
(something like 1.5gbit on a x1 link) if the pci-express
slot does not offer more bandwidth.
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Just like our other drivers before we can switch ixgbe to
provide real-time packet/byte counters to the stack easily.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Several counters behave differently on 82598 causing them to display
incorrect values. Adjust the accounting so the reported numbers
make sense and do not double count or represent the wrong item.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
We were not returning CHECKSUM_NONE in a lot of cases which is
wrong. Move common exit points in this function and error code
up before the actual work in this function.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
A gap was left in the FW release/grab code in up/down path. Fix
it by making the release/grab code a function and calling it in
appropriate locations.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
This ports Herbert Xu's "maybe_stop_tx" code and removes the tx_lock
which is not needed.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
After testing we confirmed that the irq_sem can safely be
removed from ixgbe.
Add strict state checking code to various ethtool parts to
properly protect against races between various driver reset
paths.
Signed-off-by: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
While cleaning up the internal API focussing on Fiber and CX4 code
we found that I had broken the copper PHY initialization code. This
patch restores the PHY-specific code. This is mostly uninteresting
since no copper PHY boards are yet available. The changes have been
tested against Fiber only as I do not even have copper PHY versions
of 82598 macs.
This change actually cleans up the API code a bit more and we
lose some initialization code. A few PHY link detection helper
lines of code have been snuck into this patch, as well as a
read flush where it was suspected that this might cause issues.
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
* use irq_handler_t where appropriate
* no need to use 'irq' function arg, its already stored in a data struct
* rename irq handler 'irq' argument to 'dummy', where the function
has been analyzed and proven not to use its first argument.
* remove always-false "dev_id == NULL" test from irq handlers
* remove pointless casts from void*
* declance: irq argument is not const
* add KERN_xxx printk prefix
* fix minor whitespace weirdness
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Several of the Intel ethernet drivers keep an atomic counter used to
manage when to actually hit the hardware with a disable or an enable.
The way the net_rx_work() breakout logic works during a pending
napi_disable() is that it simply unschedules the poll even if it
still has work.
This can potentially leave interrupts disabled, but that is OK
because all of the drivers are about to disable interrupts
anyways in all such code paths that do a napi_disable().
Unfortunately, this trips up the semaphore used here in the Intel
drivers. If you hit this case, when you try to bring the interface
back up it won't enable interrupts. A reload of the driver module
fixes it of course.
So what we do is make sure all the sequences now go:
napi_disable();
atomic_set(&adapter->irq_sem, 0);
*_irq_disable();
which makes sure the counter is always in the correct state.
Reported by Robert Olsson.
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes a regression added by changeset
53e52c729c ("[NET]: Make ->poll()
breakout consistent in Intel ethernet drivers.")
As pointed out by Jesse Brandeburg, for three of the drivers edited
above there is breakout logic in the *_clean_tx_irq() code to prevent
running TX reclaim forever. If this occurs, we have to elide NAPI
poll completion or else those TX events will never be serviced.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
This makes the ->poll() routines of the E100, E1000, E1000E, IXGB, and
IXGBE drivers complete ->poll() consistently.
Now they will all break out when the amount of RX work done is less
than 'budget'.
At a later time, we may want put back code to include the TX work as
well (as at least one other NAPI driver does, but by in large NAPI
drivers do not do this). But if so, it should be done consistently
across the board to all of these drivers.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Auke Kok <auke-jan.h.kok@intel.com>
Drivers do this to try to break out of the ->poll()'ing loop
when the device is being brought administratively down.
Now that we have a napi_disable() "pending" state we are going
to solve that problem generically.
Signed-off-by: David S. Miller <davem@davemloft.net>
Make strings const if possible, and fix includes so forward definitions
are seen.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
