On all of the other Intel drivers we place checksum close to TSO as they
have a significant amount in common and it can help to reduce the decision
tree for how to handle the frame as the first check in TSO is to see if
checksumming is offloaded, and if it is not we can skip _BOTH_ TSO and Tx
checksum offload based on a single check.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch is meant to rewrite the logic for how we determine if we can
transmit the frame or if it needs to be linearized.

The previous code for this function was using a mix of division and modulus
division as a part of computing if we need to take the slow path.  Instead
I have replaced this by simply working with a sliding window which will
tell us if the frame would be capable of causing a single packet to span
several descriptors.

The logic for the scan is fairly simple.  If any given group of 6 fragments
is less than gso_size - 1 then it is possible for us to have one byte
coming out of the first fragment, 6 fragments, and one or more bytes coming
out of the last fragment.  This gives us a total of 8 fragments
which exceeds what we can allow so we send such frames to be linearized.

Arguably the use of modulus might be more exact as the approach I propose
may generate some false positives.  However the likelihood of us taking much
of a hit for those false positives is fairly low, and I would rather not
add more overhead in the case where we are receiving a frame composed of 4K
pages.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

In an upcoming patch I would like to have access to the descriptor count
used for the data portion of the frame.  For this reason I am splitting up
the descriptor count function from the function that stops the ring.

Also in order to try and reduce unnecessary duplication of code I am moving
the slow-path portions of the code out of being inline calls so that we can
just jump to them and process them instead of having to build them into
each function that calls them.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch updates the code for determining the L4 protocol and L3 header
length so that when IPv6 extension headers are being used we can determine
the offset and type of the L4 protocol.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch contains a number of fixes to make certain that we are using
the correct protocols when parsing both the inner and outer headers of a
frame that is mixed between IPv4 and IPv6 for inner and outer.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Acked-by: NKiran Patil <kiran.patil@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The XL722 has support for providing the outer UDP tunnel checksum on
transmits.  Make use of this feature to support segmenting UDP tunnels with
outer checksums enabled.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This is mostly a minor clean-up for the Rx checksum path in order to avoid
some of the unnecessary conditional checks that were being applied.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Add exception handling to the Tx checksum path so that we can handle cases
of TSO where the frame is bad, or Tx checksum where we didn't recognize a
protocol

Drop I40E_TX_FLAGS_CSUM as it is unused, move the CHECKSUM_PARTIAL check
into the function itself so that we can decrease indent.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch defers writing to the Tx descriptor bits until we know we have
successfully completed a given operation.  So for example we defer updating
the tunnelling portion of the context descriptor until we have fully
identified the type.

The advantage to this approach is that we can assemble values as we go
instead of having to try and kludge everything together all at once.  As a
result we can significantly clean up the tunneling configuration for
instance as we can just do a pointer walk and do the math for the distance
between each set of points.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds support for IPv6 extension headers in setting up the Tx
checksum.  Without this patch extension headers would cause IPv6 traffic to
fail as the transport protocol could not be identified.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch fixes two issues.  First was the fact that iphdr(skb)->protocl
was being used to test for the outer transport protocol.  This completely
breaks IPv6 support.  Second was the fact that we cleared the flag for v4
going to v6, but we didn't take care of txflags going the other way.  As
such we would have the v6 flag still set even if the inner header was v4.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The Tx checksum path was maintaining a set of 3 pointers and two lengths in
order to prepare the packet for being checksummed.  The thing is we only
really needed 2 pointers, and the lengths that were being maintained can
easily be computed.

As such we can replace the IPv4 and IPv6 header pointers with one single
union that represents both, or a generic pointer to the start of the
network header.  For the L4 headers we can do the same with TCP and a
generic pointer to the start of the transport header.  The length of the
TCP header is obtained by simply multiplying doff by 4, and the network
header length can be obtained by subtracting the network header pointer
from the transport header pointer.

While I was at it I renamed l4_hdr to l4_proto to make it a bit more clear
and less likely to be confused with l4.hdr which is the transport header
pointer.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch goes through and pulls all of the spots where we were updating
either the TCP or IP checksums in the TSO and checksum path into the TSO
function.  The general idea here is that we should only be updating the
header after we verify we have completed a skb_cow_head check to verify the
head is writable.

One other advantage to doing this is that it makes things much more
obvious.  For example, in the case of IPv6 there was one spot where the
offset of the IPv4 header checksum was being updated which is obviously
incorrect.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch makes it so that the L4 header offsets and such can be ignored
when dealing with the L3 checksum and length update.  This is done making
use of two things.

First we can just use the offset from the L4 header to the start of the
packet to determine the L4 offset, and from that we can then make use of
the data offset to determine the full length of the headers.

As far as adjusting the checksum to remove the length we can simply add the
inverse of the length instead of having to recompute the entire
pseudo-header without the length.  In the case of an IPv6 header this
should be significantly cheaper since we can make use of a value we already
needed instead of having to read the source and destination address out of
the packet.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Instead of casing u32 values to u64 it makes more sense to just start out
with u64 values in the first place.  This way we don't need to create a
mess with all of the casts needed to populate a 64b value.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The i40e and i40evf drivers contained code for inserting an outer checksum
on UDP tunnels.  The issue however is that the upper levels of the stack
never requested such an offload and it results in possible errors.

In addition the same logic was being applied to the Rx side where it was
attempting to validate the outer checksum, but the logic there was
incorrect in that it was testing for the resultant sum to be equal to the
header checksum instead of being equal to 0.

Since this code is so massively flawed, and doing things that we didn't ask
for it to do I am just dropping it, and will bring it back later to use as
an offload for SKB_GSO_UDP_TUNNEL_CSUM which can make use of such a
feature.

As far as the Rx feature I am dropping it completely since it would need to
be massively expanded and applied to IPv4 and IPv6 checksums for all parts,
not just the one that supports Tx checksum offload for the outer.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The queues should never be enabled/disabled in the interrupt handler,
ICR0 interrupt enable should be the only thing that needs to be
dynamically changed in the handler.

This patch fixes that. Without this patch X722 platforms were
seeing weird ping timings when in Legacy mode since it takes
a whole lot of time for the HW/FW to re-enable queues.

Change-ID: If065afc45d81c5a19d4a94a00cd5b8f61cefc40c
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

In the case where we have a page fully used by receive data, we need to
release the page fully to the stack. Instead of calling get_page (which
increments the page count) followed by free_page (which decrements the
page count), just donate our reference to the stack. Although this
donation is not tax deductible, it does allow us to avoid two very
expensive atomic operations that reverse each other.

Change-ID: If70739792d5748995fc175ec92ac2171ed4ad8fc
Signed-off-by: NMitch Williams <mitch.a.williams@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds a workaround for cases where we might have
interrupts that got lost but WB happened.
If that happens without this patch we will see a tx_timeout.
To work around it, this patch goes ahead and reschedules NAPI
in that situation, if NAPI is not already scheduled.
We also add a counter in ethtool to keep track of when
we detect a case of tx_lost_interrupt.

Note: napi_reschedule() can be safely called from process/service_task
context and is done in other drivers as well without an issue.

Change-ID: I00f98f1ce3774524d9421227652bef20fcbd0d20
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Mr. Spock would certainly raise an eyebrow to see us using bitwise
operators, when we should clearly be relying on logic. Fascinating.

Change-ID: Ie338010c016f93e9faa2002c07c90b15134b7477
Signed-off-by: NMitch Williams <mitch.a.williams@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Refactor the packet split Rx code to properly use half-pages for
receives. The previous code was doing way more mapping and unmapping
than it needed to, and wasn't properly using half-pages.

Increment the page use count each time we give a half-page to an skb,
knowing that the stack will probably process and release the page before
we need it again. Only free and reallocate pages if the count shows that
both half-pages are in use. Add counters to track reallocations and page
reuse.

Change-ID: I534b299196036b64be82b4861a0a4036310a8f22
Signed-off-by: NMitch Williams <mitch.a.williams@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The i40e and i40evf drivers now cleanly handle allocation
failures and can avoid kernel log spew from the memory allocator
when allocations fail, so set __GFP_NOWARN on Rx buffer alloc.

Change-ID: Ic9e1b83c495e2a3ef6b069ba7fb6e52ce134cd23
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This is the "Don't Give Up" patch.  Previously the
driver could fail an allocation, and then possibly stall
a queue forever, by never coming back to continue receiving
or allocating buffers.

With this patch, the driver will keep polling trying to allocate
receive buffers until it succeeds.  This should keep all receive
queues running even in the face of memory pressure.

Also update copyright year in file header.

Change-ID: I2b103d1ce95b9831288a7222c3343ffa1988b81b
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

While re-enabling interrupts the driver would clear all pending
causes. This meant that if an interrupt was generated while the driver
was cleaning or polling with interrupts disabled, then that interrupt
was lost.  This could cause a queue to become dead, especially for
receive.  Refactored the enable_icr0 function in order to allow
it to be decided by the caller whether the CLEARPBA (clear pending
events) bit will be set while re-enabling the interrupt.

Also update copyright year in file headers.

Change-ID: Ic1db100a05e13c98919057696db147a258ca365a
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Now that the Force-WriteBack functionality in X710/XL710 devices
has been moved out of the clean routine and into the service task,
we need to make sure WriteBack-On-ITR is separated out since it
is still called from clean.

In the X722 devices, Force-WriteBack implies WriteBack-On-ITR but
without the interrupt, which put the driver into a missed
interrupt scenario and a potential tx-timeout report.

With this patch, we break the two functions out, and call the
appropriate ones at the right place. This will avoid creating missed
interrupt like scenarios for X722 devices.

Also update copyright year in file headers.

Change-ID: Iacbde39f95f332f82be8736864675052c3583a40
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The X722 can support automatic rule eviction for automatically added
flow director rules.  Feature is (should be) disabled by default.
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Don't bother trying to set up a TSO if the skb->ip_summed is not
set to CHECKSUM_PARTIAL.

Change-ID: I6495b3568e404907a2965b48cf3e2effa7c9ab55
Signed-off-by: NShannon Nelson <shannon.nelson@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Driver was using an offset based off a DMA handle while mapping and
unmapping using sync_single_range_for[cpu|device], where it should
be using DMA handle (returned from alloc_coherent) and the offset of the
memory to be sync'd.

Change-ID: I208256565b1595ff0e9171ab852de06b997917c6
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Reviewed-by: NNelson, Shannon <shannon.nelson@intel.com>
Reviewed-by: NWilliams, Mitch A <mitch.a.williams@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

We were not doing write-back on interrupt throttle for Legacy case in X722.
This patch fixes that, so we do WB_ON_ITR for Legacy as well. Plus the issue
that we should still be setting NO_ITR if we are touching the DYN_CTLN register
since we do not want to change ITR setting here.

Change-ID: I5db8491ee1544118a389db839cecc93e1bbc480e
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

intel/i40e/i40e_txrx.c: In function 'i40e_xmit_frame_ring':
intel/i40e/i40e_txrx.c:2367:20: error: 'oiph' may be used uninitialized in this function [-Werror=maybe-uninitialized]
intel/i40e/i40e_txrx.c:2317:16: note: 'oiph' was declared here
intel/i40e/i40e_txrx.c:2367:17: error: 'oudph' may be used uninitialized in this function [-Werror=maybe-uninitialized]
intel/i40e/i40e_txrx.c:2316:17: note: 'oudph' was declared here
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

If the driver calls skb_set_hash even with a zero hash, that
indicates to the stack that the hash calculation is offloaded
in hardware. So the Stack doesn't do a SW hash which is required
for load balancing if the user decides to turn of rx-hashing
on our device.

This patch fixes the path so that we do not call skb_set_hash
if the feature is disabled.

Change-ID: Ic4debfa4ff91b5a72e447348a75768ed7a2d3e1b
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds driver hooks to implement ndo_ops to add/del
udp port in the HW to identify GENEVE tunnels.
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: NKiran Patil <kiran.patil@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

1) remove duplicate include of tcp.h
2) put an ampersand at the end of a line instead of the beginning
3) remove a useless dev_info
4) match declaration of function to the implementation
5) repair incorrect comment
6) correct whitespace
7) remove unused define
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

We shouldn't be using a bitwise operator here; it's not a bitwise
operation. Use a logical operator instead. Why doesn't c have a
logical-or-and-assign operator?

Change-ID: Id84f3ca884910bed7073c84b1e16a102e958d0de
Signed-off-by: NMitch Williams <mitch.a.williams@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch fixes the memory leak which would be seen otherwise when user
programs flow-director filter using ethtool (sideband filter programming).

When ethtool is used to program flow directory filter, 'raw_buf' gets
allocated and it is supposed to be freed as part of queue cleanup. But
check of 'tx_buffer->skb' was preventing it from being freed.

Change-ID: Ief4f0a1a32a653180498bf6e987c1b4342ab8923
Signed-off-by: NKiran Patil <kiran.patil@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch contains following changes:
   - detection and recovery logic (issue SW interrupt) has been moved to
     service_task from timeout function.
   - added some more debug info from tx_timeout.

Logic to detect and recover TX queue hung is now two step process:
  - service_task detects TX queue hung and sets a bit(hung_detected) if
    it was not set.
  - if bit was set (means this is back-back hung condition detected),
    issue SW interrupt and clear the bit.
  - napi_poll clears the bit unconditionally since it cleans TX/RX queues.

Change-ID: Ieed03a48927c845a988b3ff375090bf37caeb903
Signed-off-by: NKiran Patil <kiran.patil@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Issue a prefetch for data early in the transmit path.
This should not be generally needed for Tx traffic, but
it helps immensely for pktgen workloads and should help
for forwarding workloads as well.

Change-ID: Iefee870c20599e0c4240e1d8637e4f16b625f83a
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

When in NAPI with interrupts disabled, the HW needs to be forced to do a
write back on TX if the number of descriptors pending are less than a
cache line.

This stat helps keep track of how many times we get into this situation.

Change-ID: I76c1bcc7ebccd6bffcc5aa33bfe05f2fa1c9a984
Signed-off-by: NAnjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Code was moved into a separate function some time ago.

Change-ID: Icabbe71ce05cf5d716d3e1152cdd9cd41d11bcb5
Signed-off-by: NShannon Nelson <shannon.nelson@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

We would like to automatically provide busy polling support
to all NAPI drivers, without them having to implement anything.

skb_mark_napi_id() can be called from napi_gro_receive() and
napi_get_frags().

Few drivers are still calling skb_mark_napi_id() because
they use netif_receive_skb(). They should eventually call
napi_gro_receive() instead. I will leave this to drivers
maintainers.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>