Where available, use the fine-grained counters in rtnl_link_stats64 to
indicate different RX error causes. For drop reasons, use driver-private
ethtool counters.

In particular this patch allows us to keep track of driver-side drops due
to unknown/unsupported HW descriptor format.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Instead of storing the multicast-mapped MAC address in an IP address
object, just calculate the MAC address when actually building a cmd
for the IP address.

While at it, also clean up some rather verbose copying of IP addresses.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Replace our custom implementations with the stack's version of IP address
hashing.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When setting a device online, both subdrivers have the same code to
program the HW trap and Isolation mode. Move that code into a single
place.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This consolidates the SW statistics code, and improves it to
(1) account for the header overhead of each segment on a TSO skb,
(2) count dangling packets as in-error (during eg. shutdown), and
(3) only count offloads when the skb was successfully transmitted.

We also count each segment of an TSO skb as one packet - except for
tx_dropped, to be consistent with dev->tx_dropped.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We have logic to determine the desired promisc mode in _each_ code path.
Change things around so that there is a clean split between
(a) high-level code that selects the new mode, and (b) implementations
of the various mechanisms to program this mode.

This also keeps qeth_promisc_to_bridge() from polluting the debug logs
on each RX modeset.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: NAlexandra Winter <wintera@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The cast type currently gets selected in .ndo_start_xmit, and is then
piped through several layers until it's stored into the HW header.
Push the selection down into qeth_l?_fill_header() to (1) reduce the
number of xmit-wide parameters, and (2) merge the two route validation
checks into just one.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

As follow-up to commit 0cd6783d ("s390/qeth: check dst entry before use"),
consolidate the dst_check() logic into a single helper and add a wrapper
around the cast type selection.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use napi_gro_receive() to pass up all types of packets that a L3 device
may receive.
1) For proper L2 packets received by the IQD sniffer, this is the
   obvious thing to do.
2) For af_iucv (which doesn't provide a GRO assist), the GRO code will
   transparently fall back to netif_receive_skb(). So there's no need to
   special-case this traffic in our code.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

De-duplicate the pm callback implementations from the two sub-drivers,
replacing them with core helpers that delegate to the .set_online and
.set_offline callbacks.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that all cmds are dynamically allocated, the code for static cmd
buffers can go away entirely. Resulting in a nice reduction of
code/data size & complexity, while removing the risk that
qeth_clear_cmd_buffers() releases cmds that are still in-flight.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add a new wrapper that allocates DIAG cmds of the right size, and fills
in the common fields.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch converts the adapter, assist and bridgeport cmd paths to
dynamic allocation. Most of the work is about re-organizing the cmd
headers, calculating the correct cmd length, and filling in the right
value in the sub-cmd's length field.

Since we now also set the correct length for cmds that are not reflected
by a fixed struct (ie SNMP), we can remove the work-around from
qeth_snmp_command().
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For code that uses qeth_send_simple_setassparms_prot(), we currently
can't differentiate whether the cmd should contain (1) no parameter, or
(2) a 4-byte parameter with value 0.
At the moment this doesn't cause any trouble. But when using dynamically
allocated cmds, we need to know whether to allocate & transmit an
additional 4 bytes of zeroes.
So instead of the raw parameter value, pass a parameter pointer
(or NULL) to qeth_send_simple_setassparms_prot().
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch reduces the usage of the write channel's static cmd buffers,
by dynamically allocating all simple IPA cmds (eg. STARTLAN, SETVMAC).
It also converts the OSN path.

Doing so requires some changes to how we calculate the cmd length.
Currently when building IPA cmds, we're quite generous in how much data
we send down to the device (basically the size of the biggest cmd we
know). This is no real concern at the moment, since the static cmd
buffers are backed with zeroed pages. But for dynamic allocations, the
exact length matters. So this patch also adds the needed length
calculations to each cmd path.

Commands that have multiple subtypes (eg. SETADP) of differing length
will be converted with follow-up patches.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We statically allocate 8 cmd buffers on the read channel, when the only
IO left that's still using them is the long-running READ.
Replace this with a single allocated cmd, that gets restarted whenever
the READ completed.

This introduces refcounting for allocated cmds, so that the READ cmd can
survive the IO completion.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The vast majority of SETUP-classified trace entries can be moved to
their device-specific trace file. This reduces pollution of the global
SETUP file, and provides a consistent trace view of all activity on the
device.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the tear down sequence in qeth_l?_stop_card() has finished, the
card is guaranteed to be in DOWN state and we don't have to check for
it again.
With this insight we can also remove the redundant setting of
card->state in qeth_l?_set_online()'s error path.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While qeth_l3 uses netif_keep_dst() to hold onto the dst, a skb's dst
may still have been obsoleted (via dst_dev_put()) by the time that we
end up using it. The dst then points to the loopback interface, which
means the neighbour lookup in qeth_l3_get_cast_type() determines a bogus
cast type of RTN_BROADCAST.
For IQD interfaces this causes us to place such skbs on the wrong
HW queue, resulting in TX errors.

Fix-up the various call sites to first validate the dst entry with
dst_check(), and fall back accordingly.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When selecting the cast type of a neighbourless IPv4 skb (eg. on a raw
socket), qeth_l3 falls back to the packet's destination IP address.
For this case we should classify traffic sent to 255.255.255.255 as
broadcast.
This fixes DHCP requests, which were misclassified as unicast
(and for IQD interfaces thus ended up on the wrong HW queue).
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When building the L3 HW header for non-IP packets, trust the cast type
that was passed as parameter. qeth_l3_get_cast_type() has most likely
also used h_dest to determine the cast type, so we get the same
result, and can remove that duplicated code.
In the unlikely case that we would get a _different_ cast type, then
that's based off a route lookup and should be considered authoritative.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This de-duplicates the L2 and L3 cast-type code, and makes the L2 code
a bit more robust by removing the fragile assumption that skb->data
always points to the Ethernet Header. This would break in code paths
where we pushed the HW header onto the skb.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We have helper macros for all possible device types, replace all
remaining open-coded accesses to the type fields.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Current xmit code only stops the txq after attempting to fill an
IO buffer that hasn't been TX-completed yet. In many-connection
scenarios, this can result in frequent rejected TX attempts, requeuing
of skbs with NETDEV_TX_BUSY and extra overhead.

Now that we have a proper 1-to-1 relation between stack-side txqs and
our HW Queues, overhaul the stop/wake logic so that the xmit code
stops the txq as needed.
Given that we might map multiple skbs into a single buffer, it's crucial
to ensure that the queue always provides an _entirely_ empty IO buffer.
Otherwise large skbs (eg TSO) might not fit into the last available
buffer. So whenever qeth_do_send_packet() first utilizes an _empty_
buffer, it updates & checks the used_buffers count.

This now ensures that an skb passed to qeth_xmit() can always be mapped
into an IO buffer, so remove all of the -EBUSY roll-back handling in the
TX path. We preserve the minimal safety-checks ("Is this IO buffer
really available?"), just in case some nasty future bug ever attempts to
corrupt an in-use buffer.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This adds trivial support for multiple TX queues on OSA-style devices
(both real HW and z/VM NICs). For now we expose the driver's existing
QoS mechanism via .ndo_select_queue, and adjust the number of available
TX queues when qeth_update_from_chp_desc() detects that the
HW configuration has changed.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

qeth has been supporting multiple HW Output Queues for a long time. But
rather than exposing those queues to the stack, it uses its own queue
selection logic in .ndo_start_xmit... with all the drawbacks that
entails.
Start off by switching IQD devices over to a proper mqs net_device,
and converting all the netdev_queue management code.

One oddity with IQD devices is the requirement to place all mcast
traffic on the _highest_ established HW queue. Doing so via
.ndo_select_queue seems straight-forward - but that won't work if only
some of the HW queues are active
(ie. when dev->real_num_tx_queues < dev->num_tx_queues), since
netdev_cap_txqueue() will not allow us to put skbs on the higher queues.

To make this work, we
1. let .ndo_select_queue() map all mcast traffic to netdev_queue 0, and
2. later re-map the netdev_queue and HW queue indices in
   .ndo_start_xmit and the TX completion handler.

With this patch we default to a fixed set of 1 ucast and 1 mcast queue.
Support for dynamic reconfiguration is added at a later time.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The naming of several QDIO helpers doesn't match their actual
functionality, or the structures they operate on. Clean this up.

s/qeth_alloc_qdio_buffers/qeth_alloc_qdio_queues
s/qeth_free_qdio_buffers/qeth_free_qdio_queues
s/qeth_alloc_qdio_out_buf/qeth_alloc_output_queue
s/qeth_clear_outq_buffers/qeth_drain_output_queue
s/qeth_clear_qdio_buffers/qeth_drain_output_queues
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

All users of the lock are running in process context now.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The inet6addr_chain is atomic. So instead of starting the cmd IO for
SETIP / DELIP straight from the notifier callback, run it from a
workqueue. This is the last step towards removal of cmd IO completion
polling.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Extract a little helper, so that high-level callers can manipulate the
IP table without worrying about the locking. This will make it easier
to convert the code to a different locking primitive later on.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The L2 and L3 .ndo_set_rx_mode callbacks maintain an address cache
to decide which addresses have changed since the last modeset.

When the card is set offline, qeth_l?_stop_card() drains this cache.
This happens only after 1) the net_device has been detached, and
2) any pending RX modeset has completed. Consequently we can access the
cache lock-free.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

.ndo_set_rx_mode gets called in process context, but while holding the
addr_list spinlock. Which means we currently can't sleep while
re-programming the HW, and need to poll for IO completion. That's bad,
in particular since receiving the cmd response can fail silently and
we're then polling until the timeout hits.

As a first step towards eliminating the IO completion polling, run the
RX modeset from a work element and only take the addr_list lock while
updating the RX mode address cache.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

As part of the TX completion path, qeth_release_skbs() frees the completed
skbs with __skb_queue_purge(). This ends in kfree_skb(), reporting every
completed skb as dropped.
On the other hand when dropping an skb in .ndo_start_xmit, we end up
calling consume_skb()... where we should be using kfree_skb() so that
drop monitors get notified.

Switch the drop/consume logic around, and also don't accumulate dropped
packets in the tx_errors statistics.

Fixes: dc149e37 ("s390/qeth: replace open-coded skb_queue_walk()")
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The ucast IP table is utilized by some of the L3-specific sysfs attributes
that qeth_l3_create_device_attributes() provides. So initialize the table
_before_ registering the attributes.

Fixes: ebccc739 ("s390/qeth: add missing hash table initializations")
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The HW trap and VNICC configuration is exposed via sysfs, and may have
already been modified when qeth_l?_probe_device() attempts to initialize
them. So (1) initialize the VNICC values a little earlier, and (2) don't
bother about the HW trap mode, it was already initialized before.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that qeth always uses dev_close() to shutdown the interface, we can
trust the locking and remove some custom state checks.
qeth_l?_stop_card() is no longer called for a card in UP state, so remove
the checks there too. This basically makes the UP state obsolete, so rip
out the whole thing (except for the sysfs-visible string).
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It makes no difference whether we
1. manually disarm the HW trap and call the offline code with
   recovery_mode == 1, or
2. call the offline code with recovery_mode == 0, and let it disarm the
   HW trap for us.

So consolidate the two code paths in the suspend callback.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The recovery code already runs in a kthread, we don't have to defer the
offlining further.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

smatch complains that __qeth_l3_set_offline() first accesses card->dev,
and then later checks whether the pointer is valid.
Since commit d3d1b205 ("s390/qeth: allocate netdevice early"), the
pointer is _always_ valid - that patch merely missed to remove this one
check.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When resetting an interface ("recovery"), qeth currently attempts to
elide the call to dev_close(). We initially only call .ndo_close to
quiesce the data path, and then offline & online the ccwgroup device.
If the reset succeeded, a call to .ndo_open then resumes the data path
along with some internal setup (dev_addr validation, RX modeset) that
dev_open() would have usually triggered.
dev_close() only gets called (via the close_dev worker) if the reset
action fails.

It's unclear whether this was initially done due to locking concerns, or
rather to execute the reset transparently. Either way, temporarily
closing the interface without dev_close() is fragile, and means we're
susceptible to various races and unexpected behaviour. For instance:

- Bypassing dev_deactivate_many() means that the qdiscs are not set to
__QDISC_STATE_DEACTIVATED. Consequently any intermittent TX completion
can wake up the txq, resulting in calls to .ndo_start_xmit while the
data path is down. We have custom state checking to detect this case
and drop such packets.

- Because the IFF_UP flag doesn't reflect the interface's actual state
during a reset, we have custom state checking in .ndo_open and
.ndo_close to guard against invalid calls.

- Considering that the reset might take a considerable amount of time
(in particular if an IO fails and we end up waiting for its timeout), we
_do_ want NETDEV_GOING_DOWN and NETDEV_DOWN events so that components
like bonding, team, bridge, macvlan, vlan, ... can take appropriate
action.
Signed-off-by: NJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>