This change include several changes as listed below all related to UAR.
UAR is a special PCI memory area where the so-called doorbell register and
blue flame register live. Blue flame is a feature for sending small packets
more efficiently via a PCI memory page, instead of using PCI DMA.
- All structures and functions named xxx_uuars were renamed into xxx_bfreg.
- Remove partially implemented Blueflame support from mlx5en(4) and mlx5ib.
- Implement blue flame register allocator.
- Use blue flame register allocator in mlx5ib.
- A common UAR page is now allocated by the core to support doorbell register
writes for all of mlx5en and mlx5ib, instead of allocating one UAR per
sendqueue.
- Add support for DEVX query UAR.
- Add support for 4K UAR for libmlx5.
Linux commits:
7c043e908a74ae0a935037cdd984d0cb89b2b970
2f5ff26478adaff5ed9b7ad4079d6a710b5f27e7
0b80c14f009758cefeed0edff4f9141957964211
30aa60b3bd12bd79b5324b7b595bd3446ab24b52
5fe9dec0d045437e48f112b8fa705197bd7bc3c0
0118717583cda6f4f36092853ad0345e8150b286
a6d51b68611e98f05042ada662aed5dbe3279c1e
MFC after: 1 week
Sponsored by: Mellanox Technologies // NVIDIA Networking
Use the native vnode lookup functions, instead of going via the LinuxKPI,
because the file referenced is typically created outside the LinuxKPI, and
the LinuxKPI's fdget() can only resolve file descriptor numbers which
were created by itself.
The vnode pointer is used as an identifier to identify XRCD handles which
are sharing resources.
This patch fixes the so-called XRCD support in ibcore for FreeBSD.
Refer to ibv_open_xrcd(3) for more information how the file descriptor
argument is used.
Reviewed by: kib@
MFC after: 1 week
Sponsored by: Mellanox Technologies // NVIDIA Networking
Call M_SETFIB() to make sure the IPoIB packet is directed to the correct
interface-specific FIB.
This was sufficient to allow general-purpose routing using the default FIB,
and a separate FIB for routing between IPoIB on ib0 and IPoEthernet on mce0.
Reviewed by: hselasky
Obtained from: Anmol Kumar <anmolk at panasas dot com>
MFC after: 1 week
Sponsored by: Panasas
Differential Revision: https://reviews.freebsd.org/D25239
Coverity claims the call to rdma_gid2ip in cma_igmp_send overwrites addr.
Use a consistent definition of sockaddr to prevent detections and code
changes in the future.
Submitted by: bret_ketchum@dell.com
Reported by: Coverity
Reviewed by: hselasky, kib
MFC after: 2 weeks
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D26229
Currently the linking order of the infiniband, IB, modules decide in which
order the clients are attached and detached. For example one IB client may
use resources from another IB client. This can lead to a potential deadlock
at shutdown. For example if the ipoib is unregistered after the ib_multicast
client is detached, then if ipoib is using multicast addresses a deadlock may
happen, because ib_multicast will wait for all its resources to be freed before
returning from the remove method.
Fix this by using module_xxx_order() instead of module_xxx().
Differential Revision: https://reviews.freebsd.org/D23973
MFC after: 1 week
Sponsored by: Mellanox Technologies
r357614 added CTLFLAG_NEEDGIANT to make it easier to find nodes that are
still not MPSAFE (or already are but aren’t properly marked).
Use it in preparation for a general review of all nodes.
This is non-functional change that adds annotations to SYSCTL_NODE and
SYSCTL_PROC nodes using one of the soon-to-be-required flags.
Mark all obvious cases as MPSAFE. All entries that haven't been marked
as MPSAFE before are by default marked as NEEDGIANT
Approved by: kib (mentor, blanket)
Commented by: kib, gallatin, melifaro
Differential Revision: https://reviews.freebsd.org/D23718
This reduces the number of references to VLAN_TRUNKDEV() in ibcore.
Currently only VLAN is supported as a child interface in FreeBSD.
Remove superfluous RCU locking.
Sponsored by: Mellanox Technologies
When epoch(9) was introduced to network stack, it was basically
dropped in place of existing locking, which was mutexes and
rwlocks. For the sake of performance mutex covered areas were
as small as possible, so became epoch covered areas.
However, epoch doesn't introduce any contention, it just delays
memory reclaim. So, there is no point to minimise epoch covered
areas in sense of performance. Meanwhile entering/exiting epoch
also has non-zero CPU usage, so doing this less often is a win.
Not the least is also code maintainability. In the new paradigm
we can assume that at any stage of processing a packet, we are
inside network epoch. This makes coding both input and output
path way easier.
On output path we already enter epoch quite early - in the
ip_output(), in the ip6_output().
This patch does the same for the input path. All ISR processing,
network related callouts, other ways of packet injection to the
network stack shall be performed in net_epoch. Any leaf function
that walks network configuration now asserts epoch.
Tricky part is configuration code paths - ioctls, sysctls. They
also call into leaf functions, so some need to be changed.
This patch would introduce more epoch recursions (see EPOCH_TRACE)
than we had before. They will be cleaned up separately, as several
of them aren't trivial. Note, that unlike a lock recursion the
epoch recursion is safe and just wastes a bit of resources.
Reviewed by: gallatin, hselasky, cy, adrian, kristof
Differential Revision: https://reviews.freebsd.org/D19111
When the software send queue gets filled up, callbacks to
if_transmit will stop. Make sure the transmit callback
routine checks the send queue and outputs any remaining
mbufs. Else the remaining mbufs may simply sit in the
output queue blocking the transmit path.
MFC after: 3 days
Sponsored by: Mellanox Technologies
The mistake came about like this: the first attempt to commit was blocked by
a pre-commit hook due to missing SVN tags. svn revert doesn't delete new
files, I guess. While reapplying the fixed diff, the non-empty target file
was just concatenated with the new contents? Ugh. :-(
This regression was introduced in the r326169 Linux v4.9 Infiniband upgrade.
Restore the functionality.
Reviewed by: hselasky
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D21298
This allows use of the shared _net_inet_sdp in more than one compilation
unit. (Nothing in-tree uses this today, but some of Isilon's out-of-tree
SDP enhancements add sysctls below the node.)
Sponsored by: Dell EMC Isilon
In current RDMACM implementation RDMACM server will not find a GID
index when the request was prio-tagged and the sever is non
prio-tagged and vise-versa.
According to 802.1Q-2014, VLAN tagged packets with VLAN id 0 should
be considered as untagged. Treat RDMACM request the same.
Reviewed by: hselasky, kib
MFC after: 3 Days
Sponsored by: Mellanox Technologies
This was enumerated with exhaustive search for sys/eventhandler.h includes,
cross-referenced against EVENTHANDLER_* usage with the comm(1) utility. Manual
checking was performed to avoid redundant includes in some drivers where a
common os_bsd.h (for example) included sys/eventhandler.h indirectly, but it is
possible some of these are redundant with driver-specific headers in ways I
didn't notice.
(These CUs did not show up as missing eventhandler.h in tinderbox.)
X-MFC-With: r347984
Add the new rates that were added to the Infiniband specification as part of
HDR and 2x support.
Submitted by: slavash@
MFC after: 3 days
Sponsored by: Mellanox Technologies
ib_req_notify_cq may return negative value which will indicate a
failure. In the case of uncorrectable error, we will end up in an
endless loop. Fix that, by going to another loop with poll_more
only if there is anything left to poll.
Submitted by: slavash@
MFC after: 3 days
Sponsored by: Mellanox Technologies
- Remove macros that covertly create epoch_tracker on thread stack. Such
macros a quite unsafe, e.g. will produce a buggy code if same macro is
used in embedded scopes. Explicitly declare epoch_tracker always.
- Unmask interface list IFNET_RLOCK_NOSLEEP(), interface address list
IF_ADDR_RLOCK() and interface AF specific data IF_AFDATA_RLOCK() read
locking macros to what they actually are - the net_epoch.
Keeping them as is is very misleading. They all are named FOO_RLOCK(),
while they no longer have lock semantics. Now they allow recursion and
what's more important they now no longer guarantee protection against
their companion WLOCK macros.
Note: INP_HASH_RLOCK() has same problems, but not touched by this commit.
This is non functional mechanical change. The only functionally changed
functions are ni6_addrs() and ni6_store_addrs(), where we no longer enter
epoch recursively.
Discussed with: jtl, gallatin
As it does for recv*(2), MSG_DONTWAIT indicates that the call should
not block, returning EAGAIN instead. Linux and OpenBSD both implement
this, so the change makes porting easier, especially since we do not
return EINVAL or so when unrecognized flags are specified.
Submitted by: Greg V <greg@unrelenting.technology>
Reviewed by: tuexen
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D18728
Modify QP can fail and it can be acceptable, like when moving from RST to
ERR state, all the rest are not acceptable and a message to the log
should be printed.
The current code prints on all failures and many messages like:
"Failed to modify QP to ERROR state" appear, even when supported by the
state machine of the QP object.
Linux commit:
5dc78ad1904db597bdb4427f3ead437aae86f54c
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
When a packet needs fragmentation, it might generate more than 3 fragments.
With the queue length 3, all fragments are generated faster than the
queue is drained, which effectively drops fourth and later fragments on
the floor.
Submitted by: kib@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
When changing the MTU of ibX network interfaces, check that the MTU was really
changed before requesting an update of the multicast rules. Else we might go
into an infinite loop joining and leaving ibX multicast groups towards the
opensm master interface.
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
It is not enough to set ifnet->if_mtu to change the interface MTU.
System saves the MTU for route in the radix tree, and route cache keeps
the interface MTU as well. Since addition of the multicast group causes
recalculation of MTU, even bringing the interface up changes MTU from
4042 to 1500, which makes the system configuration inconsistent. Worse,
ip_output() prefers route MTU over interface MTU, so large packets are
not fragmented and dropped on floor.
Fix it for ipoib(4) using the same approach (or hack) as was applied
for it_tun/if_tap in r339012. Thanks to bz@ for giving the hint.
Submitted by: kib@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
Binding to a loopback device is not allowed. Make sure the destination
device address is global by clearing the bound device interface.
Only do this conditionally, else link local addresses won't work.
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
A couple of places in the CM do
spin_lock_irq(&cm_id_priv->lock);
...
if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
However when the underlying transport is RoCE, this leads to a sleeping function
being called with the lock held - the callchain is
cm_alloc_response_msg() ->
ib_create_ah_from_wc() ->
ib_init_ah_from_wc() ->
rdma_addr_find_l2_eth_by_grh() ->
rdma_resolve_ip()
and rdma_resolve_ip() starts out by doing
req = kzalloc(sizeof *req, GFP_KERNEL);
not to mention rdma_addr_find_l2_eth_by_grh() doing
wait_for_completion(&ctx.comp);
to wait for the task that rdma_resolve_ip() queues up.
Fix this by moving the AH creation out of the lock.
Linux commit:
c76161181193985087cd716fdf69b5cb6cf9ee85
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
Trying to validate loopback fails because rtalloc1() resolves system
local addresses to the loopback network interface, lo0. Fix this by
explicitly checking for loopback during validation of the source
and destination network address. If the source address belongs to
a local network interface and is equal to the destination address,
there is no need to run the destination address through rtalloc1().
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
The master network interface and the VLANs may reside in different VNETs.
Make sure that all VNETs are searched when scanning for GID entries.
Submitted by: netapp
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
This prevents code from accepting RoCEv1 connections when
only ROCEv2 is enabled and vice versa.
Linux commit:
0c4386ec77cfcd0ccbdbe8c2e67dd3a49b2a4c7f
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
The array ib_mad_mgmt_class_table.method_table has MAX_MGMT_CLASS
(80) elements. Hence compare the array index with that value instead
of with IB_MGMT_MAX_METHODS (128). This patch avoids that Coverity
reports the following:
Overrunning array class->method_table of 80 8-byte elements at element index 127
(byte offset 1016) using index convert_mgmt_class(mad_hdr->mgmt_class)
(which evaluates to 127).
Linux commit:
2fe2f378dd45847d2643638c07a7658822087836
Submitted by: hselasky@
Approved by: hselasky (mentor)
MFC after: 1 week
Sponsored by: Mellanox Technologies
