On i350 when traffic is looped back from a VF to the PF the value is byte
swapped from the normal format. In order to address this we need to add a
flag indicating that the ring will need to byte swap the loopback packets
prior to processing them.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Since we mask interrupts in EIMS not in IMS there is no need to re-enable
mask bits in that register. As such we can remove the write to IMS from
the end of igb_msix_other.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change allows support for per packet timesync and global device reset
on the i350 adapter. These features were supported on both 82580 and i350
however it looks like several checks where not updated and as such the i350
support was not enabled.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change makes certain that one interrupt is always initialized in
igb_request_irq. In addition we drop the use of adapter->pdev and
instead just call pdev since we made a local copy of the pointer earlier in
the function.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This is mostly a drop of unnecessary pointer defines for q_vector when we
don't have issues with line width and don't have multiple references to
the pointer.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch cleans up several issues with VLANs on igb after the recent
changes that were meant to leave the VLANs enabled/disable via the
netdev->features flags.
Specifically the Rx VLAN settings were being dropped after reset due to the
fact that they were not being restored correctly. In addition I removed
the IRQ disable/enable since those were in place to protect the setting of
vlgrp.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Instead of doing a byte swap on the staterr bits in the Rx descriptor we can
save ourselves a bit of space and some CPU time by instead just testing for
the various bits out of the Rx descriptor directly.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Since the netdev now has its' own checksum flag to indicate if Rx checksum
is enabled we might as well use that instead of using the ring flag.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to cleanup some of the IVAR register configuration.
igb_assign_vector had become pretty large with multiple copies of the same
general code for setting the IVAR. This change consolidates most of that
code by adding the igb_write_ivar function which allows us just to compute
the index and offset and then use that information to setup the IVAR.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change moves information related to interrupt throttle rate
configuration into a separate q_vector sub-structure called a work
container. A similar change has already been made for ixgbe and this work
is based off of that.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change moves all of the ring flags into a single value. The advantage
to this is that there is one central area for all of these flags and they
can all make use of the set/test bit operations.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
There are a number of places where we have values that are stored as u16
but are being converted to int unnecessarily. In order to avoid that we
should convert all variables that deal with the next_to_clean, next_to_use,
and count to u16 values.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to update the ring and vector allocations so that they
are per node instead of allocating everything on the node that
ifconfig/modprobe is called on. By doing this we can cut down
significantly on cross node traffic.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Instead of storing most of the data for the TX hot path in the stack until
we are ready to write the descriptor we can save ourselves some time and
effort by pushing the SKB, tx_flags, gso_size, bytecount, and protocol into
the first igb_tx_buffer since that is where we will end up putting it
anyway.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change will combine the writes of tx_buffer_info and the Tx data
descriptors into a single function. The advantage of this is that we can
avoid needless memory reads from the buffer info struct and speed things up
by keeping the accesses to the local registers.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to combine all of the TX flags fields into one u32
flags field so that it can be stored into the tx_buffer_info structure.
This includes the time stamp flag as well as mapped_as_page flag info.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to cleanup the protocol handling in the transmit path
so that it correctly offloads software VLAN tagged frames.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to improve the readability of the driver by separating
out the cmd_type configuration and the olinfo configuration into their own
functions. By doing this it is much easier to determine which ingredients
go into setting up these to portions of the descriptor.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change converts two tx_buffer_info index values into pointers. The
advantage to this is that we reduce unnecessary computations and in the case
of next_to_watch we get an added bonus of the value being able to provide
additional information as a NULL value indicates it is unset versus a 0 not
having any meaning for the index value.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch is meant to simplify the transmit path by reducing the overhead
for creating a transmit context descriptor. The current implementation is
split with igb_tso and igb_tx_csum doing two separate implementations on
how to setup the tx_buffer_info structure and the tx_desc. By combining
them it is possible to reduce code and simplify things since now only one
function will create context descriptors.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
In order to be able to improve the performance of the TX path it has been
necessary to add addition info to the tx_buffer_info structure. However a
side effect is that the structure has gotten larger and this in turn has
also increased the size of the RX buffer info structure. In order to avoid
this in the future I am splitting the single buffer_info structure into two
separate ones and instead I will join them by making the buffer_info
pointer in the ring a union of the two.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is to make the NAPI budget limits for transmit
adjustable. Currently they are only set to 128, and when
the changes/improvements to NAPI occur to allow for adjustability,
it would be possible to tune the value for optimal
performance with applications such as routing.
v2: remove tie between NAPI and interrupt moderation
fix work limit define name (s/IXGBE/IGB/)
Update patch description to better reflect patch
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Instead of using the multi_tx_table to map possible Tx queues to Tx rings
we can just do simple subtraction for the unlikely event that the Tx queue
provided exceeds the number of Tx rings.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Since igb only uses advanced descriptors we might as well just use an IGB
specific define and drop the _ADV suffix for the descriptor declarations.
In addition this can be further reduced by assuming that it will be working
on pointers since that is normally how the Tx descriptors are handled.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Many of the function names in the hot path are carrying an extra "_adv"
suffix on the end of them to represent the fact that they are using
advanced descriptors instead of legacy descriptors. However since all igb
uses are advanced descriptors adding the extra suffix doesn't really add
any additional data. Since this is the case it is easiest to just drop the
suffix and save us from having to store the extra characters.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to be a general cleanup and performance improvement
for clean_rx_irq. The previous patch should have updated the allocation so
that the rings can be treated as read-only within the clean_rx_irq
function. In addition I am re-ordering the operations such that several
goals are accomplished including reducing the overhead for packet
accounting, reducing the number of items on the stack, and improving
overall performance.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to improve performance by splitting the Tx and Rx
rings into 3 sections. The first is primarily a read only section
containing basic things like the indexes, a pointer to the dev and netdev
structures, and basic information. The second section contains the stats
and next_to_use and next_to_clean values. The third section is primarily
unused values that can just be placed at the end of the ring and are not
used in the hot path.
The adapter structure has several sections that are read in the hot path.
In order to improve performance there I am combining the frequent read
hot path items into a single cache line.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to streamline the Rx buffer allocation and cleanup.
This is accomplished by reducing the number of writes by only having the Rx
descriptor ring written by software during allocation, and it will only be
read during cleanup.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change removes support for single buffer mode from igb and makes the
driver function in packet split always. The advantage to doing this is
that we can reduce total memory allocation overhead significantly as we
will only need to allocate one 1K slab per packet and then make use of a
reusable half page instead of allocating a 2K slab per packet.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch modifies the max_frame_size in order account for an optional
VLAN tag. In order to support this we must also increase the
MAX_STD_JUMBO_FRAME_SIZE to account for the 4 extra bytes.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change cleans up the RXDCTL and TXDCTL configurations and optimizes RX
performance by allowing back write-backs on all hardware other than 82576.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Aaron Brown <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch improves the memory utilization with RSC when in one-buffer
mode. This is accomplished by making the default buffer sizes match up
with the standard memory allocation sizes minus 1K for shared info and
padding overhead. By doing this CPU utilization when doing large receives
can be reduced by as much as 8%.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
The adapter structure was removed from the call so it can be dropped from
the ixgbe_fso documentation.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
One of the 82598 phys was not being correctly identified as being SFP.
This change corrects that.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change adds a small bit of missing code for enabling the overheat sensor
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
ixgbe_up and ixgbe_up_complete will always return 0. Since this doesn't
provide any useful information we might as well just make them both void
and save ourselves from having to return an unused value.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change fixes an issue in which the incorrect amount of headroom was
being reserved for flow director filters.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change fixes a minor redundancy in that tx_sample_rate was set twice.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
It was possible to inadvertently add additional interrupt causes to the
MSI-X other interrupt. This occurred when things such as RX buffer overrun
events were being triggered at the same time as an event such as a Flow
Director table reinit request. In order to avoid this we should be
explicitly programming only the interrupts that we want enabled. In
addition I am renaming the ixgbe_msix_lsc function and interrupt to drop
any implied meaning of this being a link status only interrupt.
Unfortunately the patch is a bit ugly due to the fact that ixgbe_irq_enable
needed to be moved up before ixgbe_msix_other in order to have things
defined in the correct order.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to cleanup some of the code related to SR-IOV and the
interrupt registers. Specifically I am moving the EITRSEL configuration
into the MSI-X configuration section instead of enablement. Also I am
fixing the VF shutdown path since it had operations in the incorrect order.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
The reset paths are overly complicated and are either missing steps or
contain extra unnecessary steps such as reading MAC address twice. This
change is meant to help clean up the reset paths an get things functioning
correctly.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change updated the TXDCTL configuration. The main goal is to be much
more explicit about the configuration and avoid a possible fake TX hang
when the interrupt throttle rate is set to 0.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch is a minor whitespace cleanup to compress the device ID
declaration and board type declaration onto the same line. It seems to
make sense since all of the combinations of the two are less than 80
characters and it makes the overall layout a bit more readable.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch drops a set of unnecessary dereferences to the hardware structure
since we already have a local copy of the hardware pointer.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch makes it so that the map_rings_to_vectors call will work with
all interrupt types. The advantage to this is that there will now be a
predictable mapping for all given interrupt types.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change switches us over to using the ring->dev pointer instead of
having to use the adapter->pdev->dev reference. The advantage to this is
that it is a much shorter route to get the to final needed value.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
The allocation and freeing of the IRQ affinity hint needs some updates
since there are a number of spots where we run into possible issues with
the hint not being correctly updated.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change consolidates all of the MSI-X interrupt and polling routines
into two single functions. One for the interrupt and one for the code.
The main advantage to doing this is that the compiler can optimize the
routines into single monolithic functions which should allow all of them
function to occupy a single block of memory and as such avoid jumping
around.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change makes it so that the default Tx work limit is 256 buffers or
1/2 of an entire ring instead of a full ring size so that it is much more
likely that we will be able to actually reach the work limit value.
Previously with the value set to an entire ring it would not have been
possible for us to trigger an event due to the fact that the Tx work is
stopped at the point where we cannot place one more buffer on the ring and
it is not restarted until cleanup is complete.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to fix the patch:
ixgbe: Cleanup FCOE and VLAN handling in xmit_frame_ring
And can be rolled into it if needed.
What this fixes is that VLAN tagged packets were not being tagged if they
were prio 7 which matches up with TC_PRIO_CONTROL. In order to fix it I am
just setting things up so that we always tag VLAN tagged packets.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Tested-by: Phil Schmitt <[email protected]
Signed-off-by: Jeff Kirsher <[email protected]>
This change makes it so that the CC bit in the descriptor is set when
SR-IOV is enabled. This is needed in order to support offloading
functionality when passing traffic over the internal TX switch.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change converts the current bit array into a linked list so that the
q_vectors can simply go through ring by ring and locate each ring needing
to be cleaned.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch helps to simplify the work being done by the transmit path by
removing the unnecessary compares between count and the work limit. Instead
we can simplify this by just adding a budget value that will act as a count
down from the work limit value.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to further cleanup the transmit path by streamlining
some of the VLAN and FCOE/DCB tasks in the transmit path. In addition it
adds code for support software VLANs in the event that they are used in
conjunction with DCB and/or FCOE.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
CONFIG_FCOE is not the correct define to check since it is possible for it
to be CONFIG_FCOE_MODULE, as such the reference to it should be replaced
with IXGBE_FCOE.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch implements a partial refactor of the TX map/queue and cleanup
routines. It merges the map and queue functionality and as a result
improves the transmit performance by avoiding unnecessary reads from memory.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Phil Schmitt <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch is meant to fix a memory leak found via code review for FCOE.
Specifically on DDP flows the SKBs were being dropped without being
recycled, freed, or given to the stack.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to address possible race conditions from the status
and error bits on the RX descriptors being re-read by multiple functions in
the RX cleanup path. To resolve this I have added code that will pass the
staterr value to those functions.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change moves work_limit, total_packets, and total_bytes into the ring
container struct of the q_vector. The advantage of this is that it should
reduce the size of memory used in the event of multiple rings being
assigned to a single q_vector. In addition it should help to reduce the
total workload for calculating itr since now total_packets and total_bytes
will be the total work done of the interrupt instead of for the ring.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch adds support for a ring container structure to be used within
the q_vector. The basic idea is to provide a means of separating the RX
and TX rings while maintaining a common structure for their containment.
The advantage to this is that later we should be able to pass this
structure to the update_itr functions without needing to pass individual
rings.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
The ixgbe_maybe_stop_tx function is only a few lines long and is called
multiple times through the xmit hotpath. In order to streamline things it
makes sense to just inline it.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to update ATR so that it will use the recorded RX
queue instead of the CPU in the case of routing. This change is meant to
help ixgbe default behavior to more closely match that of the kernel.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This fix makes it so that the fdir_perfect_lock is initialized in all
cases. This is necessary as the fdir_filter_exit routine will always
attempt to take the lock before inspecting the filter table.
Reported-by: Ben Greear <[email protected]>
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch updates the current methods used for determining if we have
enough space to transmit a given skb. The current method is quite wasteful
as it has us go through and determine how each page is going to be broken
up. That only needs to be done if pages are larger than our maximum data
per TXD. As such I have wrapped that in a page size check.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
There is a significant amount of shared functionality between the checksum
and TSO offload configuration that is shared in regards to how they setup
the context descriptors. Since so much of the functionality is shared it
makes sense to move the shared functionality into a single function and
just call that function from the two context descriptor specific routines.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change updates all values dealing with count, next_to_use, and
next_to_clean so that they stay u16 values. The advantage of this is that
there is no re-casting of type during the propagation through the stack.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is a minor cleanup that converts the IXGBE_DESC_UNUSED macro
into a static inline function just for the case of the code being a bit
cleaner.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change makes it so that we pass the adapter struct instead of the
netdev for most of the basic interrupts that are not associated with
q_vectors.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change fixes the fact that we would trigger a null pointer dereference
or specify the wrong ring if the rings were restored. This change makes
certain that the DROP queue is a static value, and all other rings are
based on the ring offsets for the PF.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to allow for nfc to insert and remove filters in order
to test the ethtool interface which includes it's own rules manager.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This code adds support for displaying the filters that were added via the
nfc interface. This is primarily to test the interface for now, but I am
also looking into the feasibility of moving all of the ntuple filter code
in ixgbe over to the nfc interface since it seems to be better implemented.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change adds basic support for the obtaining of RSS ring counts.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to update the internal framework of ixgbe so that
perfect filters can be stored and tracked via software.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
I am removing the requirement that Ntuple filters have the same
number of queues and requirements as ATR. As a result this change will
make it so that all the Ntuple flag does is disable ATR for now.
This change fixes an issue in which we were incorrectly re-enabling ATR
when we exited perfect filter mode. This was due to the fact that the
logic assumed RSS and DCB were mutually exclusive which is no longer the
case.
To correct this we just need to add a check to guarantee DCB is disabled
before re-enabling ATR.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Due to numerous issues in ntuple filters it has been decided to move the
interface over to the network flow classification interface. As a first
step to achieving this I first need to remove the old ntuple interface.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to remove all support for displaying an ntuple as
strings via ETHTOOL_GRXNTUPLE. The reason for this change is due to the
fact that multiple issues have been found including:
- Multiple buffer overruns for strings being displayed.
- Incorrect filters displayed, cleared filters with ring of -2 are displayed
- Setting get_rx_ntuple displays no rules if defined.
- Endianess wrong on displayed values.
- Hard limit of 1024 filters makes display functionality extremely limited
The only driver that had supported this interface was ixgbe. Since it no
longer uses the interface and due to the issues mentioned above I am
submitting this patch to remove it.
v2:
Updated based on comments from Ben Hutchings
- Left ETH_SS_NTUPLE_FILTERS in code but commented on it being deprecated
- Removed ethtool_rx_ntuple_list and ethtool_rx_ntuple_flow_spec_container
- Left ETHTOOL_GRXNTUPLE but commented it as deprecated
Also cleaned up set_rx_ntuple since there is no flow spec container to
maintain we can drop all the code for the alloc and free of it and just
return ops->set_rx_ntuple().
Signed-off-by: Alexander Duyck <[email protected]>
Acked-by: Ben Hutchings <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This patch cleans up two minor issues in ixgbe_down. Specifically it
addresses the fact that the VFs should not be pinged until after interrupts
are disabled otherwise they might still get a response. It also drops the
use of the txdctl temporary variable since the only bit we should be
writing to the TXDCTL registers during a shutdown is the flush bit.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change merges the over-temp task into the service task. As a result
all tasklets are finally combined into once single tasklet for easier
management.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change merges the ATR table reinitialization into the service task.
This is yet another opportunity to avoid any race conditions as we don't
want to be attempting to reinitialize the table during a possible reset.
In addition this change adds a counter for table reinitialization so that
it can be tracked as part of the regular statistics.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Evan Swanson <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to further help to reduce possible configuration
collisions between the various tasklets. This change combines the device
reset with the service task. As a result it is now not possible to be
updating the link on the device while also resetting the part.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Evan Swanson <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This patch is meant to merge the functionality of the ixgbe watchdog task
into the service task. By doing this all link state functionality will be
controlled by a single task. As a result the reliability of the interface
will be improved as the likelihood of any race conditions is further
reduced.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to address several race conditions with multi-speed
fiber SFP+ modules in 82599 adapters. Specifically issues have been seen
in which both the SFP configuration and the multi-speed fiber configuration
are running simultaneously which will result in the device getting into an
erroneous link down state.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Evan Swanson <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change moves flags and state into the same cacheline. The reason for
this change is because both are frequently read around the same time and
infrequently written. By combining them into the same cacheline this
should help to reduce memory utilization.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Evan Swanson <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change removes the ntuple display support from ixgbe. The reason for
this change is to resolve a number of issues in the way display filtering
is handled.
I plan to add support for displaying these filters via the network flow
classifier interface.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Evan Swanson <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to add an ntuple data extensions to the rx network flow
classification specifiers. The idea is to allow ntuple to be displayed via
the network flow classification interface.
The first patch had some left over stuff from the original flow extension
flags I had added. That bit is removed in this patch.
The second had some left over comments that stated we ignored bits in the
masks when we actually match them.
This work is based on input from Ben Hutchings.
Signed-off-by: Alexander Duyck <[email protected]>
Reviewed-by: Ben Hutchings <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This change is meant to prevent a possible null pointer dereference if
NETIF_F_NTUPLE is defined but the set_rx_ntuple function pointer is not.
The main motivation behind this patch is to eventually replace the ntuple
interfaces entirely with the network flow classifier interfaces. This
allows the device drivers to maintain the ntuple check internally while
using the network flow classifier interface for setting up and displaying
rules.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
We were incorrectly freeing IRQs that we had not requested. This change
corrects that by making certain we only free q_vectors that we have
requested IRQs for.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Tested-by: Stephen Ko <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change fixes VM pool allocation issues based on MAC address filtering,
as well as limits the scope of VF access to promiscuous mode.
Signed-off-by: Alexander Duyck <[email protected]>
Acked-by: Greg Rose <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change fixes several issues found in ntuple filtering while I was
doing the ATR refactor.
Specifically I updated the masks to work correctly with the latest version
of ethtool, I cleaned up the exception handling and added detailed error
output when a filter is rejected, and corrected several bits that were set
incorrectly in ixgbe_type.h.
The previous version of this patch included a printk that was left over from
me fixing the filter setup. This patch does not include that printk.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Stephen Ko <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This change adds a compressed input type for atr signature hash
computation. It also drops the use of the set functions when setting up
the ATR input since we can then directly setup the hash input as two dwords
that can be stored and passed as registers.
With these changes the cost of computing the has is low enough that we can
perform a hash computation on each TCP SYN flagged packet allowing us to
drop the number of flow director misses considerably in tests such as
netperf TCP_CRR.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Stephen Ko <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This change cleans up the layout of the flow director data, and the
algorithm used to calculate the hash resulting in a 35x / 3500% performance
increase versus the old flow director hash computation. The overall effect
is only a 1% increase in transactions per second though due to the fact
that only 1 packet in 20 are actually hashed upon.
TCP_RR before:
Socket Size Request Resp. Elapsed Trans.
Send Recv Size Size Time Rate
bytes Bytes bytes bytes secs. per sec
16384 87380 1 1 60.00 23059.27
16384 87380
TCP_RR after:
Socket Size Request Resp. Elapsed Trans.
Send Recv Size Size Time Rate
bytes Bytes bytes bytes secs. per sec
16384 87380 1 1 60.00 23239.98
16384 87380
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Stephen Ko <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This change forces the link down when a mailbox timeout has occurred.
Previously it was possible for a mailbox timeout to occur but for the
interface to stay up. The problem with this was that it became possible
for an interface to stay up and miss multiple requests resulting in a
possible issue since the interface will be running in an unknown state.
Signed-off-by: Alexander Duyck <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change resolves some null function pointer accesses on 82598 when a
multi-speed fiber module is inserted into the adapter.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
The q_vector back pointer was not being set in the rings so it would not
have been possible to determine the parent q_vector of the ring.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change cleans up some of the items in ixgbe_map_rings_to_vectors.
Specifically it merges the two for loops and drops the unnecessary vectors
parameter.
It also moves the vector names into the q_vectors themselves.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is meant to improve the stack utilization and simplify the math
used in ixgbe_set_itr_msix.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
There are a number of places where we use the variable j to contain the
register index of the ring. Instead of using such a non-descriptive
variable name it is better that we name it reg_idx so that it is clear what
the variable contains.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
This change is just to cleanup some confusing logic in ixgbe_cache_ring_rss
which can be simplified by adding a conditional with return to the start of
the call.
Signed-off-by: Alexander Duyck <[email protected]>
Tested-by: Ross Brattain <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>
Alexander Duyck