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>
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>
Commit 1b524963 ("IPoIB/cm: Use common CQ for CM send completions")
changed how the high-order bits of work request IDs were used, which
had the effect that IPOIB_CM_RX_DRAIN_WRID was no longer handled as a
connected mode receive completion. This leads to the messages
ib1: cm send completion event with wrid 1073741823 (> 64)
ib1: RX drain timing out
when an interface with connected mode QPs is brought down. Fix this
by making sure that both IPOIB_OP_CM and IPOIB_OP_RECV are set in
IPOIB_CM_RX_DRAIN_WRID.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Use the same CQ for CM send completions as for all other IPoIB
completions. This means all completions are processed via the same
NAPI polling routine. This should help reduce the number of
interrupts for bi-directional traffic (such as TCP) and fixes "driver
is hogging interrupts" errors reported for IPoIB send side, e.g.
<https://bugs.openfabrics.org/show_bug.cgi?id=508>
To do this, keep a per-interface counter of outstanding send WRs, and
stop the interface when this counter reaches the send queue size to
avoid CQ overruns.
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
It's too hard to figure out what "!likely(...)" really means, and who
knows how compilers interpret the hint.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Use the stats member of struct netdevice in IPoIB, so we can save
memory by deleting the stats member of struct ipoib_dev_priv, and save
code by deleting ipoib_get_stats().
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make the way QP is being created in ipoib_cm_create_tx_qp()
consistent with ipoib_cm_create_rx_qp().
Signed-off-by: Dotan Barak <dotanb@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The IB CM should include the HCA ACK delay when calculating the local
ACK timeout value to use for RC QPs. If the HCA ACK delay is large
enough relative to the packet life time, then if it is not taken into
account, the calculated timeout value ends up being too small, which
can result in "retry exceeded" errors.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Fix
drivers/infiniband/ulp/ipoib/ipoib_cm.c:1151: warning: unused variable 'dev'
by getting rid of the variable dev, which is only used if CONFIG_IPV6
is enabled, and replacing the one use of it with the value it is
assigned, namely priv->dev.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
If a page can't be allocated for the frag list of a skb, the code to
unmap the partially allocated list is off by one. For exaple, if
'frags' equals one, i == 0, and the alloc_page() fails, then the old
loop would have unmapped mapping[1] which is uninitialized. The same
would happen if the call to ib_dma_map_page() failed.
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Acked-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
IPoIB connected mode currently rejects a connection request unless the
supported MTU is >= the local netdevice MTU. This breaks
interoperability with implementations that might have tweaked
IPOIB_CM_MTU, and there's real no longer a reason to do so: this test
is just a leftover from when we did not tweak MTU per-connection. Fix
this by making the test as permissive as possible.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Fix a crasher bug in IPoIB CM: once a QP is in the RTR state, a
receive completion (or even an asynchronous error) might be observed
on this QP, so we have to initialize all of our receive data
structures before moving to the RTR state.
As an optimization (since modify_qp might take a long time), the
jiffies update done when moving RX to the passive_ids list is also
left in place to reduce the chance of the RX being misdetected as
stale.
This fixes bug <https://bugs.openfabrics.org/show_bug.cgi?id=662>.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
commit 518b1646 ("IPoIB/cm: Fix SRQ WR leak") introduced a severe
performance regression on Mellanox cards, because keeping a QP in the
error state for extended periods of time moves hardware to the slow
path (until the QP is destroyed). For example, MPI latency goes from
~3 usecs to ~7 usecs.
Fix this by posting a send WR on one of the QPs that are being
flushed, instead of using a separate drain QP that is kept in the
error state.
This fixes bug <https://bugs.openfabrics.org/show_bug.cgi?id=636>,
reported and bisected by Scott Weitzenkamp at Cisco and debugged by
Sasha Mikheev at Voltaire.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Since NAPI polling is disabled while ipoib_cm_dev_stop() is running,
ipoib_cm_dev_stop() must poll the CQ itself in order to see the
packets draining.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
time_after() was used backwards, so the timeout occurred immediately.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
SRQ WR leakage has been observed with IPoIB/CM: e.g. flipping ports on
and off will, with time, leak out all WRs and then all connections
will start getting RNR NAKs. Fix this in the way suggested by spec:
move the QP being destroyed to the error state, wait for "Last WQE
Reached" event and then post WR on a "drain QP" connected to the same
CQ. Once we observe a completion on the drain QP, it's safe to call
ib_destroy_qp.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
In the presence of some running RX connections, we repeat
queue_delayed_work calls each 4 RX WRs, which is a waste. It's enough
to start stale task when a first passive connection is added, and
rerun it every IPOIB_CM_RX_DELAY as long as there are outstanding
passive connections.
This removes some code from RX data path.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Convert the IP-over-InfiniBand network device driver over to using
NAPI to handle completions for the main CQ. This covers all receives
as well as datagram mode sends; send completions for connected mode
connections are still handled from interrupt context.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Add a num_comp_vectors member to struct ib_device and extend
ib_create_cq() to pass in a comp_vector parameter -- this parallels
the userspace libibverbs API. Update all hardware drivers to set
num_comp_vectors to 1 and have all ULPs pass 0 for the comp_vector
value. Pass the value of num_comp_vectors to userspace rather than
hard-coding a value of 1.
We want multiple CQ event vector support (via MSI-X or similar for
adapters that can generate multiple interrupts), but it's not clear
how many vectors we want, or how we want to deal with policy issues
such as how to decide which vector to use or how to set up interrupt
affinity. This patch is useful for experimenting, since no core
changes will be necessary when updating a driver to support multiple
vectors, and we know that we want to make at least these changes
anyway.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The IPoIB CM spec allows the use of a single connection in both
active->passive and passive->active directions. The current Linux
code uses one connection for both directions, but if another node only
uses one connection for both directions, we oops when we try to look
up the passive connection. Fix by checking that qp_context is
non-NULL before dereferencing it.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
If skb allocation fails when we start the device, we call
ipoib_cm_dev_stop() even though ipoib_cm_dev_open() did not run to
completion, so we pass an invalid pointer to ib_destroy_cm_id and get
an oops.
Fix by clearing cm.id on error, and testing it in cm_dev_stop().
This fixes <https://bugs.openfabrics.org/show_bug.cgi?id=561>
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
For the common, open coded 'skb->mac.raw = skb->data' operation, so that we can
later turn skb->mac.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple case, next will handle the slightly more
"complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are quite a few places in ipoib_cm.c where we know IRQs are
enabled because we do something that sleeps in the same function, so
we can convert several occurrences of spin_lock_irqsave() to a plain
spin_lock_irq(). This cleans up the source a little and makes the
code smaller too:
add/remove: 0/0 grow/shrink: 1/5 up/down: 3/-51 (-48)
function old new delta
ipoib_cm_tx_reap 403 406 +3
ipoib_cm_stale_task 146 145 -1
ipoib_cm_dev_stop 173 172 -1
ipoib_cm_tx_handler 964 956 -8
ipoib_cm_rx_handler 956 937 -19
ipoib_cm_skb_reap 212 190 -22
Signed-off-by: Roland Dreier <rolandd@cisco.com>
There's no point in printing the opcode field in the completion
handling debugging output, since the type of completion is already
printed at the beginning of the line. In fact the opcode field is not
even defined for completions with a status other than success.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Receive buffers need to be mapped with DMA_FROM_DEVICE. Incorrectly
mapping with DMA_TO_DEVICE causes a hard lock on ppc64 machines with
an IOMMU.
This fixes <https://bugs.openfabrics.org/show_bug.cgi?id=431>
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The packet length checks in ipoib are broken: we add 4 bytes (IPoIB
encapsulation header) when sending a packet, not 20 bytes (hardware
address length) to each packet. Therefore, if connected mode is
enabled so that the interface MTU is larger than the multicast MTU,
IPoIB may end up trying to send too-long multicast packets. For
example, multicast is broken if a message of size 2048 bytes is sent
on an interface with UD MTU 2048, because 2048 is bigger than the real
limit of 2044 but the code tests against the wrong limit of 2060.
This patch fixes <https://bugs.openfabrics.org/show_bug.cgi?id=418>,
submitted by Scott Weitzenkamp <sweitzen@cisco.com>.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The sense of the time_after_eq() test in ipoib_cm_stale_task() is
reversed so that only non-stale connections are reaped. Fix this by
changing to time_before_eq().
Noticed by Pradeep Satyanarayana <pradeep@us.ibm.com>.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Avoid the overhead of freeing/reallocating and mapping/unmapping for
DMA pages that have not been written to by hardware.
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
ipoib_cm_alloc_rx_skb() might be called from IRQ context, so it must
use dev_kfree_skb_any(), not kfree_skb().
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The following patch adds experimental support for IPoIB connected
mode, as defined by the draft from the IETF ipoib working group. The
idea is to increase performance by increasing the MTU from the maximum
of 2K (theoretically 4K) supported by IPoIB on top of UD. With this
code, I'm able to get 800MByte/sec or more with netperf without
options on a Mellanox 4x back-to-back DDR system.
Some notes on code:
1. SRQ is used for scalability to large cluster sizes
2. Only RC connections are used (UC does not support SRQ now)
3. Retry count is set to 0 since spec draft warns against retries
4. Each connection is used for data transfers in only 1 direction, so
each connection is either active(TX) or passive (RX). 2 sides that
want to communicate create 2 connections.
5. Each active (TX) connection has a separate CQ for send completions -
this keeps the code simple without CQ resize and other tricks
6. To detect stale passive side connections (where the remote side is
down), we keep an LRU list of passive connections (updated once per
second per connection) and destroy a connection after it has been
unused for several seconds. The LRU rule makes it possible to avoid
scanning connections that have recently been active.
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Pradeep Satyanarayana