This patch replaces the existing mechanism for determining the correct
value to program for adaptive ITR with yet another new and more
complicated approach.

The basic idea from a 30K foot view is that this new approach will push the
Rx interrupt moderation up so that by default it starts in low latency and
is gradually pushed up into a higher latency setup as long as doing so
increases the number of packets processed, if the number of packets drops
to 4 to 1 per packet we will reset and just base our ITR on the size of the
packets being received. For Tx we leave it floating at a high interrupt
delay and do not pull it down unless we start processing more than 112
packets per interrupt. If we start exceeding that we will cut our interrupt
rates in half until we are back below 112.

The side effect of these patches are that we will be processing more
packets per interrupt. This is both a good and a bad thing as it means we
will not be blocking processing in the case of things like pktgen and XDP,
but we will also be consuming a bit more CPU in the cases of things such as
network throughput tests using netperf.

One delta from this versus the ixgbe version of the changes is that I have
made the interrupt moderation a bit more aggressive when we are in bulk
mode by moving our "goldilocks zone" up from 48 to 96 to 56 to 112. The
main motivation behind moving this is to address the fact that we need to
update less frequently, and have more fine grained control due to the
separate Tx and Rx ITR times.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch is mostly prep-work for replacing the current approach to
programming the dynamic aka adaptive ITR. Specifically here what we are
doing is splitting the Tx and Rx ITR each into two separate values.

The first value current_itr represents the current value of the register.

The second value target_itr represents the desired value of the register.

The general plan by doing this is to allow for deferring the update of the
ITR value under certain circumstances. For now we will work with what we
have, but in the future I hope to change the behavior so that we always
only update one ITR at a time using some simple logic to determine which
ITR requires an update.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

While testing code for the recent ITR changes I found that updating the Tx
ITR appeared to have no effect with everything defaulting to the Rx ITR. A
bit of digging narrowed it down the fact that we were asking the PF to
associate all causes with ITR 0 as we weren't populating the itr_idx values
for either Rx or Tx.

To correct it I have added the configuration for these values to this
patch. In addition I did some minor clean-up to just add a local pointer
for the vector map instead of dereferencing it based off of the index
repeatedly. In my opinion this makes the resultant code a bit more readable
and saves us a few characters.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Instead of using the register value for the defines when setting up the
ring ITR we can just use the actual values and avoid the use of shifts and
macros to translate between the values we have and the values we want.

This helps to make the code more readable as we can quickly translate from
one value to the other.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The CLEARPBA bit in the dynamic interrupt control register actually has
no effect either way on the hardware. As per errata 28 in the XL710
specification update the interrupt is actually cleared any time the
register is written with the INTENA_MSK bit set to 0. As such the act of
toggling the enable bit actually will trigger the interrupt being
cleared and could lead to potential lost events if auto-masking is
not enabled.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch is a further clean-up related to the change over to using
q_vector->reg_idx when accessing the ITR registers. Specifically the code
appears to have several other spots where we were computing the register
offset manually and this resulted in errors in a few spots.

Specifically in the i40evf functions for mapping queues to vectors it
appears we may have had an off by 1 error since (v_idx - 1) for the first
q_vector with an index of 0 would result in us returning -1 if I am not
mistaken.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Currently in i40e_set_priv_flags we use new_flags to check for the
I40E_FLAG_DISABLE_FW_LLDP flag.  This is an issue for a few a reasons.
DISABLE_FW_LLDP is persistent across reboots/driver reloads.  This means
we need some way to detect if FW LLDP is enabled on init.  We do this by
trying to init_dcb and if it fails with EPERM we know LLDP is disabled
in FW.

This could be a problem on older FW versions or NPAR enabled PFs because
there are situations where the FW could disable LLDP, but they do _not_
support using this flag to change it.  If we do end up in this
situation, the flag will be set, then when the user tries to change any
priv flags, the driver thinks the user is trying to disable FW LLDP on a
FW that doesn't support it and essentially forbids any priv flag
changes.

The fix is simple, instead of checking if this flag is set, we should be
checking if the user is trying to _change_ the flag on unsupported FW
versions.

This patch also adds a comment explaining that the cmpxchg is the point
of no return.  Once we put the new flags into pf->flags we can't back
out.
Signed-off-by: NAlan Brady <alan.brady@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds a warning message when the link-down-on-close flag is
setting on. The warning is printed only on MFP devices
Signed-off-by: NPaweł Jabłoński <pawel.jablonski@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds necessary delay for 4.33 firmware to recover after
EMP reset. Without this patch driver occasionally reinitializes
structures too quickly to communicate with firmware after EMP reset
causing AdminQ to timeout.
Signed-off-by: NFilip Sadowski <filip.sadowski@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The logic for dynamic ITR update is confusing at best as there were odd
paths chosen for how to find the rings associated with a given queue based
on the vector index and other inconsistencies throughout the code.

This patch is an attempt to clean up the logic so that we can more easily
understand what is going on. Specifically if there is a Rx or Tx ring that
is enabled in dynamic mode on the q_vector it is allowed to override the
other side of the interrupt moderation. While it isn't correct all this
patch is doing is cleaning up the logic for now so that when we come
through and fix it we can more easily identify that this is wrong.

The other big change made here is that we replace references to:
	vsi->rx_rings[q_vector->v_idx]->itr_setting
with:
	q_vector->rx.ring->itr_setting

The general idea is we can avoid the long pointer chase since just
accessing q_vector->rx.ring is a single pointer access versus having to
chase down vsi->rx_rings, and then finding the pointer in the array, and
finally chasing down the itr_setting from there.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The rings are already split out into Tx and Rx rings so it doesn't make
sense to have any single ring store both a Tx and Rx itr_setting value.
Since that is the case drop the pair in favor of storing just a single ITR
value.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

'bufer' should be 'buffer'
Signed-off-by: NAlan Brady <alan.brady@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Having these checks in ibmvnic_xmit causes problems with VLAN
tagging and balance-alb/tlb bonding modes. The restriction they
imposed can be removed.
Signed-off-by: NJohn Allen <jallen@linux.vnet.ibm.com>
Signed-off-by: NNathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For dwmac4, GMAC_INT_DEFAULT_ENABLE already includes
GMAC_INT_PMT_EN, so it is redundant to check if hw->pmt
is set, and if so, setting the bit again.

For dwmac1000, GMAC_INT_DEFAULT_MASK does not include
GMAC_INT_DISABLE_PMT, so it is redundant to check if
hw->pmt is set, and if so, clearing an already cleared bit.

Improve code readability by removing this redundant code.
Signed-off-by: NNiklas Cassel <niklas.cassel@axis.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

GMAC_INT_DEFAULT_MASK is written to the interrupt enable register.
In previous versions of the IP (e.g. dwmac1000), this register was
instead an interrupt mask register.
To improve clarity and reflect reality, rename GMAC_INT_DEFAULT_MASK
to GMAC_INT_DEFAULT_ENABLE.
Signed-off-by: NNiklas Cassel <niklas.cassel@axis.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The interrupt status register in both dwmac1000 and dwmac4 ignores
interrupt enable (for dwmac4) / interrupt mask (for dwmac1000).
Therefore, if we want to check only the bits that can actually trigger
an irq, we have to filter the interrupt status register manually.

Commit 0a764db1 ("stmmac: Discard masked flags in interrupt status
register") fixed this for dwmac1000. Fix the same issue for dwmac4.

Just like commit 0a764db1 ("stmmac: Discard masked flags in
interrupt status register"), this makes sure that we do not get
spurious link up/link down prints.
Signed-off-by: NNiklas Cassel <niklas.cassel@axis.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When allocating RX or TX buffer pools, the driver needs to provide a
unique mapping ID to firmware for each pool. This value is assigned
using a counter which is incremented after a new pool is created. The
ID can be an integer ranging from 1-255. When migrating to a device
that requests a different number of queues, this value was not being
reset properly. As a result, after enough migrations, the counter
exceeded the upper bound and pool creation failed. This is fixed by
resetting the counter to one in this case.
Signed-off-by: NThomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The Cavium thunder nicvf driver supports rx/tx rings of up to 65536 entries per.
The number of entires are stored in the q_len member of struct q_desc_mem. The
problem is that q_len being a u16, results in 65536 becoming 0.

In getting pointers to descriptors in the rings, the driver uses q_len minus 1
as a mask after incrementing the pointer, in order to go back to the beginning
and not go past the end of the ring.

With the q_len set to 0 the mask is no longer correct and the driver does go
beyond the end of the ring, causing various ills. Usually the first thing that
shows up is a "NETDEV WATCHDOG: enP2p1s0f1 (nicvf): transmit queue 7 timed out"
warning.

This patch remedies the problem by changing q_len to a u32.
Signed-off-by: NDean Nelson <dnelson@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While handling a driver reset we get a H_CLOSED return trying
to send a CRQ event. When this occurs we need to queue up another
reset attempt. Without doing this we see instances where the driver
is left in a closed state because the reset failed and there is no
further attempts to reset the driver.
Signed-off-by: NNathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

DKMS and similar out-of-tree module replacement services use
module version to make sure the out-of-tree software is not
older than the module shipped with the kernel.  We use the
kernel version in ethtool -i output, put it into MODULE_VERSION
as well.
Reported-by: NJan Gutter <jan.gutter@netronome.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NDirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Most FWs limit the number of TSO segments a frame can produce
to 64.  This is for fairness and efficiency (of FW datapath)
reasons.  If a frame with larger number of segments is submitted
the FW will drop it.
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

All netdevs which can accept TC offloads must implement
.ndo_set_features().  nfp_reprs currently do not do that, which
means hw-tc-offload can be turned on and off even when offloads
are active.

Whether the offloads are active is really a question to nfp_ports,
so remove the per-app tc_busy callback indirection thing, and
simply count the number of offloaded items in nfp_port structure.

Fixes: 8a276873 ("nfp: provide infrastructure for offloading flower based TC filters")
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Tested-by: NPieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

nfp_port is a structure which represents an ASIC port, both
PCIe vNIC (on a PF or a VF) or the external MAC port.  vNIC
netdev (struct nfp_net) and pure representor netdev (struct
nfp_repr) both have a pointer to this structure.  nfp_reprs
always have a port associated.  nfp_nets, however, only represent
a device port in legacy mode, where they are considered the
MAC port. In switchdev mode they are just the CPU's side of
the PCIe link.

By definition TC offloads only apply to device ports.  Don't
set the flag on vNICs without a port (i.e. in switchdev mode).
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Tested-by: NPieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Upcoming changes will require all netdevs supporting TC offloads
to have a full struct nfp_port.  Require those for BPF offload.
The operation without management FW reporting information about
Ethernet ports is something we only support for very old and very
basic NIC firmwares anyway.
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Tested-by: NPieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Immed relocation is missing a shift which means for larger
offsets the lower and higher part of the address would be
ORed together.

Fixes: ce4ebfd8 ("nfp: bpf: add helpers for updating immediate instructions")
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NJiong Wang <jiong.wang@netronome.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>

It was discovered that simple program which indefinitely sends 200b UDP
packets and runs on TI AM574x SoC (SMP) under RT Kernel triggers network
watchdog timeout in TI CPSW driver (<6 hours run). The network watchdog
timeout is triggered due to race between cpsw_ndo_start_xmit() and
cpsw_tx_handler() [NAPI]

cpsw_ndo_start_xmit()
	if (unlikely(!cpdma_check_free_tx_desc(txch))) {
		txq = netdev_get_tx_queue(ndev, q_idx);
		netif_tx_stop_queue(txq);

^^ as per [1] barier has to be used after set_bit() otherwise new value
might not be visible to other cpus
	}

cpsw_tx_handler()
	if (unlikely(netif_tx_queue_stopped(txq)))
		netif_tx_wake_queue(txq);

and when it happens ndev TX queue became disabled forever while driver's HW
TX queue is empty.

Fix this, by adding smp_mb__after_atomic() after netif_tx_stop_queue()
calls and double check for free TX descriptors after stopping ndev TX queue
- if there are free TX descriptors wake up ndev TX queue.

[1] https://www.kernel.org/doc/html/latest/core-api/atomic_ops.htmlSigned-off-by: NGrygorii Strashko <grygorii.strashko@ti.com>
Reviewed-by: NIvan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This change will guard against a double free in the case that the
buffers were previously freed at some other time, such as during
a device reset. It resolves a kernel oops that occurred when changing
the VNIC device's MTU.
Signed-off-by: NThomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

At some point, a check was added to exit the polling routine during resets.
This makes sense for most reset conditions, but for a non-fatal error, we
expect the polling routine to continue running to properly clean up the rx
queues. This patch checks if we are performing a non-fatal reset and if we
are, continues normal polling operation.
Signed-off-by: NJohn Allen <jallen@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix the number of queues per enabled TC and report available queues
to the kernel without having to limit them to the max RSS limit so
they are available to be mapped for XPS. This allows a queue per
processing thread available for handling traffic for the given
traffic class.
Signed-off-by: NAmritha Nambiar <amritha.nambiar@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add the appropriate SPDX license tags to the Sun network drivers
as outlined in Documentation/process/license-rules.rst.
Signed-off-by: NShannon Nelson <shannon.nelson@oracle.com>
Reviewed-by: NZhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit baf50868 ("cxgb4: restructure VF mgmt code") has reordered
some code but an error handling label has not been updated accordingly.
So fix it and free 'adapter' if 't4_wait_dev_ready()' fails.

Fixes: baf50868 ("cxgb4: restructure VF mgmt code")
Signed-off-by: NChristophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If the driver receives a TX CQE with status as 0x1 or 0x9 or 0xb,
the completion indexes should not be used. The driver must stop
consuming CQEs from this TXQ/CQ. The TXQ from this point on-wards
to be in a bad state. Driver should destroy and recreate the TXQ.

0x1: LANCER_TX_COMP_LSO_ERR
0x9 LANCER_TX_COMP_SGE_ERR
0xb: LANCER_TX_COMP_PARITY_ERR

Reset the adapter if driver sees this error in TX completion. Also
adding sge error counter in ethtool stats.
Signed-off-by: NSuresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Lancer HW cannot handle a TSO packet with a single segment.
Disable TSO/GSO for such packets.
Signed-off-by: NSuresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit 84ce5b98 ("scripts: kernel-doc: improve nested logic to
handle multiple identifiers") improved the handling of nested structure
definitions in scripts/kernel-doc, and changed the expected format of
documentation.  This causes new warnings to appear on W=1 builds.

Only comment changes.
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Due to a typo, the mask was destroyed by a comparison instead of a bit
shift.
Signed-off-by: NWolfram Sang <wsa+renesas@sang-engineering.com>
Acked-by: NTom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch makes sure that the firmware version is never NULL. Moreover,
it also performs some cleanup on the error messages.

Fixes: a107311d ("ibmvnic: fix firmware version when no firmware level
has been provided by the VIOS server")
Signed-off-by: NDesnes A. Nunes do Rosario <desnesn@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The Mediatek ethernet driver fails to build after commit 23c35f48
("pinctrl: remove include file from <linux/device.h>") because it relies
on the pinctrl/consumer.h and pinctrl/devinfo.h being pulled in by the
device.h header implicitly.

Include these headers explicitly to avoid the build failure.

Cc: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: NThierry Reding <treding@nvidia.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Currently the async EQ has 256 entries only. It might not be big enough
for the SW to handle all the needed pending events. For example, in case
of many QPs (let's say 1024) connected to a SRQ created using NVMeOF target
and the target goes down, the FW will raise 1024 "last WQE reached" events
and may cause EQ overrun. Increase the EQ to more reasonable size, that beyond
it the FW should be able to delay the event and raise it later on using internal
backpressure mechanism.
Signed-off-by: NMax Gurtovoy <maxg@mellanox.com>
Reviewed-by: NSagi Grimberg <sagi@grimberg.me>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

gcc-8 points out that the skb_copy_to_linear_data() argument points to
the skb itself, which makes it run into a problem with overlapping
memcpy arguments:

In file included from include/linux/ip.h:20,
                 from drivers/net/ethernet/qlogic/qlge/qlge_main.c:26:
drivers/net/ethernet/qlogic/qlge/qlge_main.c: In function 'ql_realign_skb':
include/linux/skbuff.h:3378:2: error: 'memcpy' source argument is the same as destination [-Werror=restrict]
  memcpy(skb->data, from, len);

It's unclear to me what the best solution is, maybe it ought to use a
different helper that adjusts the skb data in a safe way. Simply using
memmove() here seems like the easiest workaround.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

passing the strlen() of the source string as the destination
length is pointless, and gcc-8 now warns about it:

drivers/net/ethernet/qlogic/qed/qed_debug.c: In function 'qed_grc_dump':
include/linux/string.h:253: error: 'strncpy' specified bound depends on the length of the source argument [-Werror=stringop-overflow=]

This changes qed_grc_dump_big_ram() to instead uses the length of
the destination buffer, and use strscpy() to guarantee nul-termination.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>