Connected mode is now tested and used by lots of people. No need to
hide it under CONFIG_EXPERIMENTAL.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Print the return code of ib_sa_path_rec_get() if it fails to help
debug errors.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Now that we have a specific lock to protect the network
device unicast and multicast lists, remove extraneous
grabs of the TX lock in cases where the code only needs
address list protection.
Signed-off-by: David S. Miller <davem@davemloft.net>
Add netif_addr_{lock,unlock}{,_bh}() helpers.
Use them to protect operations that operate on or read
the network device unicast and multicast address lists.
Also use them in cases where the code simply wants to
block calls into the driver's ->set_rx_mode() and
->set_multicast_list() methods.
Signed-off-by: David S. Miller <davem@davemloft.net>
Increase IPoIB ring sizes to twice their original sizes (RX: 128->256,
TX: 64->128) to act as a shock absorber for high traffic peaks. With
the current settings, we have seen cases that there are many calls to
netif_stop_queue(), which causes degradation in throughput. Also,
larger receive buffer sizes help IPoIB in CM mode to avoid experiencing
RNR NAK conditions due to insufficient receive buffers at the SRQ.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Since IPoIB connected mode does not NETIF_F_SG, we only have one DMA
mapping per send, so we don't need a mapping[] array. Define a new
struct with a single u64 mapping member and use it for the CM tx_ring.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
When the driver sets the MTU of the net device outside of its
change_mtu method, it should make use of dev_set_mtu() instead of
directly setting the mtu field of struct netdevice. Otherwise
functions registered to be called upon MTU change will not get called
(this is done through call_netdevice_notifiers() in dev_set_mtu()).
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Use of this lock is required to synchronize changes to the netdvice's
data structs. Also move the call to ipoib_flush_paths() after the
modification of the netdevice flags in set_mode().
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
ipoib_mcast_detach() does nothing except call ib_detach_mcast(), so just
use the core API in the one place that does a multicast group detach.
add/remove: 0/1 grow/shrink: 0/1 up/down: 0/-105 (-105)
function old new delta
ipoib_mcast_leave 357 319 -38
ipoib_mcast_detach 67 - -67
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The current code will set the Q_Key for any join of a non-sendonly
multicast group. The operation involves a modify QP operation, which
is fairly heavyweight, and is only really required after the join of
the broadcast group. Fix this by adding a parameter to ipoib_mcast_attach()
to control when the Q_Key is set.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
No need for a mutex around calls to ib_attach_mcast/ib_detach_mcast
since these operations are synchronized at the HW driver layer.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The IPOIB_MCAST_STARTED flag is not used at all since commit b3e2749b
("IPoIB: Don't drop multicast sends when they can be queued"), so
remove it.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The patch tries to solve the problem of device going down and paths being
flushed on an SM change event. The method is to mark the paths as candidates for
refresh (by setting the new valid flag to 0), and wait for an ARP
probe a new path record query.
The solution requires a different and less intrusive handling of SM
change event. For that, the second argument of the flush function
changes its meaning from a boolean flag to a level. In most cases, SM
failover doesn't cause LID change so traffic won't stop. In the rare
cases of LID change, the remote host (the one that hadn't changed its
LID) will lose connectivity until paths are refreshed. This is no
worse than the current state. In fact, preventing the device from
going down saves packets that otherwise would be lost.
Signed-off-by: Moni Levy <monil@voltaire.com>
Signed-off-by: Moni Shoua <monis@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Add "ipoib_use_lro" module parameter to enable LRO and an
"ipoib_lro_max_aggr" module parameter to set the max number of packets
to be aggregated. Make LRO controllable and LRO statistics accessible
through ethtool.
Signed-off-by: Vladimir Sokolovsky <vlad@mellanox.co.il>
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Set IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK for IPoIB's UD QPs if
supported by the underlying device. This creates an improvement of up
to 39% in bandwidth when sending multicast packets with IPoIB, and an
improvment of 12% in cpu usage.
Signed-off-by: Ron Livne <ronli@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
For devices that don't support SRQs, ipoib_cm_post_receive_nonsrq() is
called from both ipoib_cm_handle_rx_wc() and ipoib_cm_nonsrq_init_rx(),
and these two callers are not synchronized against each other.
However, ipoib_cm_post_receive_nonsrq() always reuses the same receive
work request and scatter list structures, so multiple callers can end
up stepping on each other, which leads to posting garbled work
requests.
Fix this by having the caller pass in the ib_recv_wr and ib_sge
structures to use, and allocating new local structures in
ipoib_cm_nonsrq_init_rx().
Based on a patch by Pradeep Satyanarayana <pradeep@us.ibm.com> and
David Wilder <dwilder@us.ibm.com>, with debugging help from Hoang-Nam
Nguyen <hnguyen@de.ibm.com>.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The connected mode implementation in the IPoIB driver has a large
overhead in the way SKBs are handled in the receive flow. It usually
allocates an SKB with as big as was used in the currently received SKB
and moves unused fragments from the old SKB to the new one. This
involves a loop on all the remaining fragments and incurs overhead on
the CPU. This patch, for small SKBs, allocates an SKB just large
enough to contain the received data and copies to it the data from the
received SKB. The newly allocated SKB is passed to the stack and the
old SKB is reposted.
When running netperf, UDP small messages, without this pach I get:
UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
14.4.3.178 (14.4.3.178) port 0 AF_INET
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # 10^6bits/sec
114688 128 10.00 5142034 0 526.31
114688 10.00 1130489 115.71
With this patch I get both send and receive at ~315 mbps.
The reason that send performance actually slows down is as follows:
When using this patch, the overhead of the CPU for handling RX packets
is dramatically reduced. As a result, we do not experience RNR NAK
messages from the receiver which cause the connection to be closed and
reopened again; when the patch is not used, the receiver cannot handle
the packets fast enough so there is less time to post new buffers and
hence the mentioned RNR NACKs. So what happens is that the
application *thinks* it posted a certain number of packets for
transmission but these packets are flushed and do not really get
transmitted. Since the connection gets opened and closed many times,
each time netperf gets the CPU time that otherwise would have been
given to IPoIB to actually transmit the packets. This can be verified
when looking at the port counters -- the output of ifconfig and the
oputput of netperf (this is for the case without the patch):
tx packets
==========
port counter: 1,543,996
ifconfig: 1,581,426
netperf: 5,142,034
rx packets
==========
netperf 1,1304,089
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
We saw a kernel oops in our regression testing when a multicast "join
finish" occurred just after the interface was -- this is
<https://bugs.openfabrics.org/show_bug.cgi?id=1040>. The test
randomly causes the HCA physical port to go down then up.
The cause of this is that ipoib_mcast_join_finish() processing happen
just after ipoib_mcast_dev_flush() was invoked (in which case the
broadcast pointer is NULL). This patch tests for and handles the case
where priv->broadcast is NULL.
Cc: <stable@kernel.org>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Commit f56bcd80 ("IPoIB: Use separate CQ for UD send completions")
introduced a bug where the transmit queue could get stopped and never
woken up. The problem is that send completions are only polled at the
end of the xmit function, so if the send queue fills up and the xmit
path stops the queue, then there is no way for send completions to
ever get polled, and so the transmit queue stays stopped forever.
Fix this by arming the send CQ just before posting the last send
request that fills the send queue. Then, when the completion event
handler is called, drain the send CQ. Since it is possible that not
enough send completions are in the CQ, verify that the the net queue
has been woken up after draining the send CQ, and if not arm a timer
and drain again at the timer function.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
When creating a child interface, copy the MTU information from the
parent. Otherwise when the child's multicast join completes, the MTU
will not be updated since the code does
dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
and priv->admin_mtu will be set to 0.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Use a dedicated CQ for UD send completions. Also, do not arm the UD
send CQ, which reduces the number of interrupts generated. This patch
farther reduces overhead by not calling poll CQ for every posted send
WR -- it does polls only when there 16 or more outstanding work requests.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
This patch enables IPoIB to use 4K UD messages (when the underlying
device and fabrics support a 4K MTU) by using two scatter buffers when
PAGE_SIZE is less than or equal to thhe HCA IB MTU size. The first
buffer is for IPoIB header + GRH header, and the second buffer is the
IPoIB payload, which is 4K-4.
Signed-off-by: Shirley Ma <xma@us.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
If a P_Key is deleted and then re-added at the same index, then IPoIB
gets confused because __ipoib_ib_dev_flush() only checks whether the
index is the same without checking whether the P_Key was present, so
the interface is stopped when the P_Key is deleted, but the event when
the P_Key is re-added gets ignored and the interface never gets
restarted.
Also, switch to using ib_find_pkey() instead of ib_find_cached_pkey()
everywhere in IPoIB, since none of the places that look for P_Keys are
in a fast path or in non-sleeping context, and in general we want to
kill off the whole caching infrastructure eventually. This also fixes
consistency problems caused because some IPoIB queries were cached and
some were uncached during the window where the cache was not updated.
Thanks to Venkata Subramonyam <vsubramo@cisco.com> for debugging this
problem and testing this fix.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
This can be used to tune at run time the parameters controlling the
event (interrupt) generation rate and thus reduce the overhead
incurred by handling interrupts resulting in better throughput. Since
IPoIB uses a single CQ for both RX and TX, RX is chosen to dictate
configuration for both RX and TX.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Just add the infrastructure so we can add functionality later.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
For HCAs that support TCP segmentation offload (IB_DEVICE_UD_TSO), set
NETIF_F_TSO and use HW LSO to offload TCP segmentation.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Convert list_splice() + INIT_LIST_HEAD() to the equivalent list_splice_init()
Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
For HCAs that support checksum offload (ie that set IB_DEVICE_UD_IP_CSUM
in the device capabilities flags), have IPoIB set NETIF_F_IP_CSUM and
use the HCA to generate and verify IP checksums.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Commit 7143740d ("IPoIB: Add send gather support") made struct
ipoib_tx_buf significantly larger, since the mapping member changed
from a single u64 to an array with MAX_SKB_FRAGS + 1 entries. This
means that allocating tx_rings with kzalloc() may fail because there
is not enough contiguous memory for the new, much bigger size. Fix
this regression by allocating the rings with vmalloc() instead.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Commit 7143740d ("IPoIB: Add send gather support") made it possible
for tx_wr.num_sge to be != 1 -- this happens if send gather support is
enabled. However, the code in the connected mode post_send() function
assumes the old invariant, namely that tx_wr.num_sge is always 1. Fix
this by explicitly setting tx_wr.num_sge to 1 in the CM post_send().
Signed-off-by: Roland Dreier <rolandd@cisco.com>
When set_multicast_list() is called the multicast task is restarted
and the IPOIB_MCAST_STARTED bit is cleared. As a result for some
window of time, multicast packets are not transmitted nor queued but
rather dropped by ipoib_mcast_send(). These dropped packets are
painful in two cases:
- bonding fail-over which both calls set_multicast_list() on the new
active slave and sends Gratuitous ARP through that slave.
- IP_DROP_MEMBERSHIP code which both calls set_multicast_list() on the
device and issues IGMP leave.
In both these cases, depending on the scheduling of the IPoIB
multicast task, the packets would be dropped. As a result, in the
bonding case, the failover would not be detected by the peers until
their neighbour is renewed the neighbour (which takes a few tens of
seconds). In the IGMP case, the IP router doesn't get an IGMP leave
and would only learn on that from further probes on the group (also a
delay of at least a few tens of seconds).
Fix this by allowing transmission (or queuing) depending on the
IPOIB_FLAG_OPER_UP flag instead of the IPOIB_MCAST_STARTED flag.
Signed-off-by: Olga Shern <olgas@voltaire.com>
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Commit efcd9971 ("IPoIB/cm: Factor out ipoib_cm_free_rx_reap_list()")
introduced a bug in ipoib_cm_dev_stop() when the receive drain times
out. In that case, the function moves all the pending rx stuff into a
private list but then calls ipoib_cm_free_rx_reap_list(), which
handles a different list.
Fix this by moving everything to the rx_reap_list that will actually
get freed up.
This fixes <https://bugs.openfabrics.org/show_bug.cgi?id=906>.
Signed-off-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
struct ipoib_cm_tx.ibwc is unused since commit 1b524963 ("IPoIB/cm:
Use common CQ for CM send completions"), so remove it.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
In P_Key event handling, if the old P_Key is no longer available, the
driver must call ipoib_ib_dev_stop() -- just as it does when the P_Key
is still available (see procedure __ipoib_ib_dev_flush()).
When a P_Key becomes available, the driver will perform ipoib_open(),
which assumes that the QP is in RESET, the cm_id has been
destroyed/deleted, etc. If ipoib_ib_dev_stop() is not called as
described above, then these assumptions will be false, and the attempt
to bring the interface up will fail.
Found by Mellanox QA.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
This patch acts as a preparation for using checksum offload for IB
devices capable of inserting/verifying checksum in IP packets. The
patch does not actaully turn on NETIF_F_SG - we defer that to the
patches adding checksum offload capabilities.
We only add support for send gathers for datagram mode, since existing
HW does not support checksum offload on connected QPs.
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
All current InfiniBand devices can handle all DMA addresses, and it's
hard to imagine anyone would be silly enough to build a new device
that couldn't. Therefore, enable the NETIF_F_HIGHDMA feature for IPoIB.
This has no effect for no, but is needed when we enable gather/scatter
support and checksum stateless offloads.
Signed-off-by: Eli Cohen <eli@mellnaox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Commit 732a2170 ("IB/ipoib: Bound the net device to the ipoib_neigh
structue") left a misleading debug print (n->dev would be a bond
device only if boding is used). Clean it up.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Move up the code that checks for a situation where the remote GID
stored in the ipoib_neigh is different than the one present in the
neighbour (handle gratuitous ARP) or that a bonding fail over has
happened but the neighbour still has a pointer to an ipoib_neigh
created by a different device than the current slave. This will cause
the driver to apply the check also for connected mode neighbours.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
qdisc_run() now tests for queue_stopped() before calling
__qdisc_run(), and the same check is done in every iteration of
__qdisc_run(), so another check is not required in the driver xmit.
This means that ipoib_start_xmit() no longer needs to test
netif_queue_stopped(); the test was added to fix earlier kernels,
where the networking stack did not guarantee that the xmit method of
an LLTX driver would not be called after the queue was stopped, but
current kernels do provide this guarantee.
To validate, I put a debug in the TX_BUSY path which never hit with 64
threads running overnight exercising this code a few 100 million
times.
Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Some HCAs (such as ehca2) support SRQ, but only support fewer than 16 SG
entries for SRQs. Currently IPoIB/CM implicitly assumes all HCAs will
support 16 SG entries for SRQs (to handle a 64K MTU with 4K pages). This
patch removes that restriction by limiting the maximum MTU in connected
mode to what the maximum number of SRQ SG entries allows.
This patch addresses <https://bugs.openfabrics.org/show_bug.cgi?id=728>
Signed-off-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
An IPoIB subnet on an IB fabric that spans multiple IB subnets can't
use link-local scope in multicast GIDs. The existing routines that
map IP/IPv6 multicast addresses into IB link-level addresses hard-code
the scope to link-local, and they also leave the partition key field
uninitialised. This patch adds a parameter (the link-level broadcast
address) to the mapping routines, allowing them to initialise both the
scope and the P_Key appropriately, and fixes up the call sites.
The next step will be to add a way to configure the scope for an IPoIB
interface.
Signed-off-by: Rolf Manderscheid <rvm@obsidianresearch.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Some IB adapters (notably IBM's eHCA) do not implement SRQs (shared
receive queues). The current IPoIB connected mode support only works
on devices that support SRQs.
Fix this by adding support for using the receive queue of each
connected mode receive QP. The disadvantage of this compared to using
an SRQ is that it means a full queue of receives must be posted for
each remote connected mode peer, which means that total memory usage
is potentially much higher than when using SRQs. To manage this, add
a new module parameter "max_nonsrq_conn_qp" that limits the number of
connections allowed per interface.
The rest of the changes are fairly straightforward: we use a table of
struct ipoib_cm_rx to hold all the active connections, and put the
table index of the connection in the high bits of receive WR IDs.
This is needed because we cannot rely on the struct ib_wc.qp field for
non-SRQ receive completions. Most of the rest of the changes just
test whether or not an SRQ is available, and post receives or find
received packets in the right place depending on the answer.
Cleaning up dead connections actually becomes simpler, because we do
not have to do the "last WQE reached" dance that is required to
destroy QPs attached to an SRQ. We just move the QP to the error
state and wait for all pending receives to be flushed.
Signed-off-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
[ Completely rewritten and split up, based on Pradeep's work. Several
bugs fixed and no doubt several bugs introduced. - Roland ]
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Factor out the code for going through the rx_reap list of struct
ipoib_cm_rx and freeing each one. This consolidates the code
duplicated between ipoib_cm_dev_stop() and ipoib_cm_rx_reap() and
reduces the risk of error when adding additional accounting.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Factor out the code to create an SRQ and allocate the receive ring in
ipoib_cm_dev_init() into a new function ipoib_cm_create_srq(). This
will make the code neater when support for devices that don't implement
SRQs is added.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Factor out the code to unmap/free skbs and free the receive ring in
ipoib_cm_dev_cleanup() into a new function ipoib_cm_free_rx_ring().
This function will be called from a couple of other places when
support for devices that don't implement SRQs is added.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
If a port goes down, ipoib_ib_dev_down() is invoked -- which flushes
the mcasts (clearing priv->broadcast) and clearing the path record
cache. If ipoib_start_xmit() is then invoked (before the broadcast
group is rejoined), a kernel oops results from attempting to access
priv->broadcast, which is still unset.
Returning NULL from path_rec_create() if priv->broadcast is NULL is a
harmless way of bypassing the problem -- the offending packet is
simply discarded "without prejudice."
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
