The Solarflare 8000 series NIC will use a new TSO scheme. The current
driver refuses to load if the current TSO scheme is not found. Remove
that check and instead make the TSO version a per-queue parameter.
Signed-off-by: NBert Kenward <bkenward@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

NAPI drivers no longer need to observe a particular protocol
to benefit from busy polling (CONFIG_NET_RX_BUSY_POLL=y)

napi_hash_add() and napi_hash_del() are automatically called
from core networking stack, respectively from
netif_napi_add() and netif_napi_del()

This patch depends on free_netdev() and netif_napi_del() being
called from process context, which seems to be the norm.

Drivers might still prefer to call napi_hash_del() on their
own, since they might combine all the rcu grace periods into
a single one, knowing their NAPI structures lifetime, while
core networking stack has no idea of a possible combining.

Once this patch proves to not bring serious regressions,
we will cleanup drivers to either remove napi_hash_del()
or provide appropriate rcu grace periods combining.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

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>

This pci_error_handlers structure is never modified, like all the other
pci_error_handlers structures, so declare it as const.

Done with the help of Coccinelle.
Signed-off-by: NJulia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

dma_set_mask already checks for a supported DMA mask before updating it,
the call to dma_supported is redundant.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Cc: Solarflare linux maintainers <linux-net-drivers@solarflare.com>
Cc: Shradha Shah <sshah@solarflare.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When the IP stack passes SKBs the sfc driver puts them in 2 different TX
queues (called partners), one for checksummed and one for not checksummed.
If the SKB has xmit_more set the driver will delay pushing the work to the
NIC.

When later it does decide to push the buffers this patch ensures it also
pushes the partner queue, if that also has any delayed work. Before this
fix the work in the partner queue would be left for a long time and cause
a netdev watchdog.

Fixes: 70b33fb0 ("sfc: add support for skb->xmit_more")
Reported-by: NJianlin Shi <jishi@redhat.com>
Signed-off-by: NMartin Habets <mhabets@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch reduces the overhead of locking for busy poll.
Previously the state was protected by a lock, whereas now
it's manipulated solely with atomic operations.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

On EF10, MC_CMD_VPORT_RECONFIGURE can cause a CODE_MC_REBOOT event
to be sent to a function without incrementing the (adapter-wide)
warm_boot_count.  In this case, the reboot is not detected by the
loop on efx_mcdi_poll_reboot(), so prepare for recovery from an MC
reboot anyway.  When this codepath is run, the MC has always just
rebooted, so this recovery is valid.

The loop on efx_mcdi_poll_reboot() is still required for other MC
reboot cases, so that actions in response to an MC reboot are
performed, such as clearing locally calculated statistics.
Siena NICs are unaffected by this change as the above scenario
does not apply.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The sfc driver internally uses a time format based on 32-bit (unsigned)
seconds and 32-bit nanoseconds. This means it will overflow in 2106,
but the value we pass into it is a signed 32-bit tv_sec that already
overflows in 2038 to a negative value.

This patch changes the logic to use the lower 32 bits of the timespec64
tv_sec in efx_ptp_ns_to_s_ns, which will have the correct value beyond the overflow.
While this does not change any of the register values, it lets us
keep using the driver after we deprecate the use of the timespec type
in the kernel.

In the efx_ptp_process_times function, the change to use timespec64
is similar, in that the tv_sec portion is ignored anyway and we only
care about the nanosecond portion that remains unchanged.
Acked-by: NRichard Cochran <richardcochran@gmail.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Reviewed-by: NThomas Gleixner <tglx@linutronix.de>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NJohn Stultz <john.stultz@linaro.org>

The pps_event_time uses two 'timespec' structures internally, which
suffer from the y2038 problem. The uses of this structure are
fairly self-contained in the pps code, so this replaces them all at
once.

Unfortunately, this includes the sfc ethernet driver aside from the
pps subsystem, so we change that one as well. Both touch the
same data structure, and there probably is no good way to split
the patch into smaller units.
Acked-by: NRichard Cochran <richardcochran@gmail.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Reviewed-by: NThomas Gleixner <tglx@linutronix.de>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NJohn Stultz <john.stultz@linaro.org>

After commit:

  654672d4 ("locking/atomics: Add _{acquire|release|relaxed}() variants of some atomic operations")

Architectures may only provide {cmp,}xchg_relaxed definitions in
asm/cmpxchg.h. Other variants, such as {cmp,}xchg, may be built in
linux/atomic.h, which means simply including asm/cmpxchg.h may not get
the definitions of all the{cmp,}xchg variants.

Therefore, we should privatize the inclusions of asm/cmpxchg.h to
keep it only included in arch/* and replace the inclusions outside
with linux/atomic.h
Signed-off-by: NBoqun Feng <boqun.feng@gmail.com>
Signed-off-by: NPeter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: NWill Deacon <will.deacon@arm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Aybuke Ozdemir <aybuke.147@gmail.com>
Cc: Chris Brannon <chris@the-brannons.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirk Reiser <kirk@reisers.ca>
Cc: Kishon Vijay Abraham I <kishon@ti.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Samuel Thibault <samuel.thibault@ens-lyon.org>
Cc: Shradha Shah <sshah@solarflare.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: William Hubbs <w.d.hubbs@gmail.com>
Cc: devel@driverdev.osuosl.org
Cc: linux-net-drivers@solarflare.com
Cc: speakup@linux-speakup.org
Link: http://lkml.kernel.org/r/1440589966-26280-1-git-send-email-boqun.feng@gmail.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

Previously, the driver would refuse to load if it couldn't secure
enough VIs from the MC to fulfill its RSS requirements.
This was causing probe to fail on later functions in
configurations where we'd run out of VIs, such as having many
VFs.

This change allows the driver to load with fewer VIs, down to a
minimum of 2. A warning will be printed saying that RSS
requirements were not met, possibly affecting performance.

efx->max_tx_channels needs to be set to avoid going down the
failure path in efx_probe_nic() immediately in the loop after the
probe() NIC-type function.
Also, Set rc=ENOSPC when bombing out of efx_probe_nic due to lack
of VIs.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Some of the stats handling code differs based on SR-IOV support,
and SRIOV support is only available if full-featured firmware is
used.
Do not use vadaptor stats if firmware mode is not set to
full-featured.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The policy in the net driver is to attempt MCDI commands and
then handle any EPERM error codes appropriately when returned
by unprivileged functions.
The ethtool selftest contains some tests which are useful on
an unprivileged function, such as the event queue interrupt
tests, but other tests cannot be performed as the function
does not have the required permissions.

If a test returns -EPERM, act as though the test was not run
and continue.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Separate functions for inserting individual and promisc filters; explicit
 fallback logic in efx_ef10_filter_sync_rx_mode(), in order not to overload
 the 'promisc' flag as also meaning "fall back to promisc".
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If the workaround to support cascaded multicast filters ("workaround_26807") is
enabled, the broadcast filter and individual multicast filters are not inserted
when in promiscuous or allmulti mode.

There is a race while inserting and removing filters when entering and leaving
promiscuous mode.  When changing promiscuous state with cascaded multicast
filters, the old multicast filters are removed before inserting the new filters
to avoid duplicating packets; this can lead to dropped packets until all
filters have been inserted.

The efx_nic:mc_promisc flag is added to record the presence of a multicast
promiscuous filter; this gives a simple way to tell if the promiscuous state is
changing.
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This change is only re-factoring; there are no changes to functionality
 except for a slight elaboration of an error message (on mismatch filter
 insertion failure).
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If a function is in promiscuous mode and another function has a broadcast or
 multicast filter inserted, the function in promiscuous mode won't see that
 broadcast or multicast traffic.
Most notably this breaks broadcast, which means ARP doesn't work. Less
 show-stoppingly, a function listening on a multicast address that's also in
 promiscuous mode will not see that multicast traffic if another function is
 also listening on that multicast address.
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When enabling the workaround for cascaded multicast filters, the MC
 can reset other functions if they have already inserted filters.
 In that case, the workaround has been enabled, but print an info
 message in the log recording that other functions had to be reset.

As other functions were reset, the MC will have incremented its boot
 count, so also increment the warm_boot_count on the function which
 enabled the workaround, as that function won't have received an MC
 reboot event and does not need to reset.
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The initial use of this will be to check a flag reporting if an FLR was
performed on other functions when enabling cascaded multicast filters.
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

GET_WORKAROUNDS was only introduced in May 2014, not all firmware
 will have it.  So call sites need to handle ENOSYS.
In this case we're probing the bug26807 workaround, which is not
 implemented in any firmware that doesn't have GET_WORKAROUNDS.
 So interpret ENOSYS as 'false'.
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After creating event queue 0, check to see if the workaround is enabled,
 and enable it if necessary.  This will be called during PCI probe and
 also when coming back up after a reset.  The nic_data->workaround_26807
 will be used in the future to control the filter insertion behaviour
 based on this workaround.

Only the primary PF can enable this workaround, so tolerate an EPERM
 error and continue.  Otherwise, if any step in the checking and enabling
 of the workaround fails, the event queue must be removed.

We check that workaround is implemented before trying to enable it,
 and store the current workaround setting before trying to change it.
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

CC: Solarflare linux maintainers <linux-net-drivers@solarflare.com>
CC: Shradha Shah <sshah@solarflare.com>
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Acked-by: NRichard Cochran <richardcochran@gmail.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The limit for BQL is updated each time we call
netdev_tx_completed_queue.
Without this patch the BQL limit was updated for every TX event we
see.
The issue was that this only updated the limit to handle the data
we complete in two events as the first event wouldn't show that
enough traffic had been processed between them.

This was OK when interrupt moderation was off but not when it was
on as more data had to be completed in a single interrupt.

The patch changes this so that we do report the completion to BQL
only when all the TX events in the interrupt have been processed.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch avoids the double up_write to filter_sem if
efx_net_open() fails.

Resolves: 2d432f20Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Some versions of MCFW do not support the MC_CMD_VADAPTOR_SET_MAC
command, and ENOSYS will be returned.

If the PF created its own vport, the function's datapath must be
stopped and the vport can be reconfigured to reflect the new MAC
address.

If the MCFW created the vport for the PF (which is the case when
the nic_data->vport_mac is blank), nothing further needs to be
done as the vport is not under the control of the PF.

This only applies to PFs because the MCFW in question does not
support VFs.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Re-organize the structure of error handling to avoid having
to duplicate the netif_err() around the ifdefs.

The only change to the behaviour of the error-handling is that
the PF's data structure to record VF details should only be
updated if the original command succeeded.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Without this change, modprobe -r sfc hits the BUG_ON() in
efx_pci_remove_main().

Fixes: e7fef9b4 ("sfc: add sysfs entry to control MCDI tracing")
Reported-by: NJarod Wilson <jarod@redhat.com>
Reviewed-by: NJarod Wilson <jarod@redhat.com>
Signed-off-by: NEdward Cree <ecree@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If any VF is assigned as the PF is unloaded, do not attempt to
remove its vport or the vswitch.  These will be removed if the
driver binds to the PF again, as an entity reset occurs during
probe.

A 'force' flag is added to efx_ef10_pci_sriov_disable() to
distinguish between disabling SR-IOV and driver unload.
SR-IOV cannot be disabled if VFs are assigned to guests.

If the PF driver is unloaded while VFs are assigned, the driver
may try to bind to the VF again at a later point if the driver
has been reloaded and the VF returns to the same domain as the PF.
In this case, the PF will not have a VF data structure, so the VF
can check this and drop out of probe early.

In this case, efx->vf_count will be zero but VFs will be present.
The user is advised to remove the VF and re-create it. The check
at the beginning of efx_ef10_pci_sriov_disable() that
efx->vf_count is non-zero is removed to allow SR-IOV to be
disabled in this case. Also, if the PF driver is unloaded, it
will disable SR-IOV to remove these unknown VFs.

By not disabling bus-mastering if VFs are still assigned, the VF
will continue to pass traffic after the PF has been removed.

When using the max_vfs module parameter, if VFs are already
present do not try to initialise any more.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the driver unloads, force the unbind and removal of any
VFs in the host with the PF.  The PF cannot remove vports and
vswitches if they are still being used by a VF driver, and when
unloading the sfc driver the removal order is not guaranteed,
so the instruction from the PF to the VF to unbind enforces a
suitable ordering so that vswitches and vports can be removed.

As a result of this, manually unbinding the driver from a single
PF will result in all of its VFs in the host also being removed.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The ifenslave command to set up a bond runs in an atomic
context, and it queries the stats on the devices that are
being enslaved. A VF needs to make an MCDI call to update
its stats, which is not allowed in atomic context.

The releasing of the stats_lock is moved to the beginning of
the VF stats update function so that in_interrupt() can be
used; it must be taken again before returning from this
function.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The raw_mask array is not initialised, so it needs to be
explicitly set to zero in the 'else' branch.

If the EVB capability is not present, a port cannot have multiple
functions so the per-port MAC stats are correct and should match
the corresponding vadaptor stats, so this redundancy can be
removed from the ethtool stats output.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

MC_CMD_MAC_STATS can be called on a function before a
vadaptor has been created, as the kernel can call into this
through ndo_get_stats/ndo_get_stats64.

If MC_CMD_MAC_STATS is called before the DMA queues have been
setup, so that a vadaptor has not been created yet, firmware
will return ENOENT. This is expected, so suppress the MCDI
error message in this case.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The netdevice statistics (in /proc/net/dev) are per-function
stats so they must use the vadaptor stats. Change the use of
MAC stats to vadaptor stats, and remove any statistics that
can only be measured per-port.  All stats that are removed
will be shown as zeroes when these statistics are displayed.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Firmware does not support a periodic DMA of vadaptor-stats
on VFs, so only update the stats buffer when stats are
requested (when running "ethtool -S" or an ip/ifconfig
command that reports stats).
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

All interfaces will display vadaptor statistics, so set all the
relevant bits in the stats bitmask. Only functions with the
LINKCTRL flag will see other stats, including (per-port) MAC stats.

The vadaptor stats are from rx_unicast to tx_overflow.
Signed-off-by: NShradha Shah <sshah@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>