Current implementation for MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK is not
supported when link layer is Ethernet.
This patch will add counter based implementation for multicast loopback
prevention. HW can drop multicast loopback packets if sender QP counter
index is equal to receiver QP counter index. If qp flag
MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK is set and link layer is Ethernet,
create a new counter and attach it to the QP so it will continue
receiving multicast loopback traffic but it's own.
The decision if to create a new counter is being made at the qp
modification to RTR after the QP's port is set. When QP is destroyed or
moved back to reset state, delete the counter.
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This is an infrastructure step for allocating and attaching more than
one counter to QPs on the same port. Allocate a counters table and
manage the insertion and removals of the counters in load and unload of
mlx4 IB.
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Manage RoCE gid table with logic in IB/core, which is common to all
vendors, and remove the mechanism from the mlx4 IB driver.
Since management of the GID cache may lead to index mismatch with the
hardware GID table, a translation between indexes is required when
modifying a QP or creating an address handle.
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This is an infrastructure step to attach all the QPs opened from the
IB driver to a counter in order to collect VF stats from the PF using
those counters.
If the port's type is Ethernet, the counter policy demands two counters
per port (one for RoCE and one for Ethernet). The port default counter
(allocated in mlx4_core) is used for the Ethernet netdev QPs and we
allocate another counter for RoCE.
If the port's traffic is Infiniband, the counter policy demands
one counter per port, so it can use the port's default counter.
Also, Add 'allocated' flag for each counter in order to clean it at
unload.
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reserve the last valid counter index for "sink" counter, when a
new counter cannot be allocated, the driver will use this counter.
In order to avoid allocating this counter on any other flow, fix the
indices bitmap allocation range, and reserve the sink counter index.
Add macro for the sink counter index and replace all appearences of the
index with the macro.
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current code decreases from the mss size (which is the gso_size
from the kernel skb) the size of the packet headers.
It shouldn't do that because the mss that comes from the stack
(e.g IPoIB) includes only the tcp payload without the headers.
The result is indication to the HW that each packet that the HW sends
is smaller than what it could be, and too many packets will be sent
for big messages.
An easy way to demonstrate one more aspect of the problem is by
configuring the ipoib mtu to be less than 2*hlen (2*56) and then
run app sending big TCP messages. This will tell the HW to send packets
with giant (negative value which under unsigned arithmetics becomes
a huge positive one) length and the QP moves to SQE state.
Fixes: b832be1e40 ('IB/mlx4: Add IPoIB LSO support')
Reported-by: Matthew Finlay <matt@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Pull InfiniBand/RDMA updates from Roland Dreier:
- Re-enable on-demand paging changes with stable ABI
- Fairly large set of ocrdma HW driver fixes
- Some qib HW driver fixes
- Other miscellaneous changes
* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (43 commits)
IB/qib: Add blank line after declaration
IB/qib: Fix checkpatch warnings
IB/mlx5: Enable the ODP capability query verb
IB/core: Add on demand paging caps to ib_uverbs_ex_query_device
IB/core: Add support for extended query device caps
RDMA/cxgb4: Don't hang threads forever waiting on WR replies
RDMA/ocrdma: Fix off by one in ocrdma_query_gid()
RDMA/ocrdma: Use unsigned for bit index
RDMA/ocrdma: Help gcc generate better code for ocrdma_srq_toggle_bit
RDMA/ocrdma: Update the ocrdma module version string
RDMA/ocrdma: set vlan present bit for user AH
RDMA/ocrdma: remove reference of ocrdma_dev out of ocrdma_qp structure
RDMA/ocrdma: Add support for interrupt moderation
RDMA/ocrdma: Honor return value of ocrdma_resolve_dmac
RDMA/ocrdma: Allow expansion of the SQ CQEs via buddy CQ expansion of the QP
RDMA/ocrdma: Discontinue support of RDMA-READ-WITH-INVALIDATE
RDMA/ocrdma: Host crash on destroying device resources
RDMA/ocrdma: Report correct state in ibv_query_qp
RDMA/ocrdma: Debugfs enhancments for ocrdma driver
RDMA/ocrdma: Report correct count of interrupt vectors while registering ocrdma device
...
In case handle_eth_ud_smac_index fails, we need to free the allocated resources.
Fixes: 2f5bb47368 ("mlx4: Add ref counting to port MAC table for RoCE")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The driver exposes interfaces that directly relate to HW state. Upon fatal
error, consumers of these interfaces (ULPs) that rely on completion of
all their posted work-request could hang, thereby introducing dependencies
in shutdown order. To prevent this from happening, we manage the
relevant resources (CQs, QPs) that are used by the device. Upon a fatal error,
we now generate simulated completions for outstanding WQEs that were not
completed at the time the HW was reset.
It includes invoking the completion event handler for all involved CQs so that
the ULPs will poll those CQs. When polled we return simulated CQEs with
IB_WC_WR_FLUSH_ERR return code enabling ULPs to clean up their resources and
not wait forever for completions upon receiving remove_one.
The above change requires an extra check in the data path to make sure that when
device is in error state, the simulated CQEs will be returned and no further
WQEs will be posted.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the mlx4 IB (RoCE) device works in link aggregation mode, it
exposes a single port to upper layers. Therefore, applications always
set '1' in port_num attribute when modifying a QP or creating an address handle.
To make sure that a node uses all available ports the mlx4 driver will
override the port_num attribute with a round robin policy.
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A0 hybrid steering is a form of high performance flow steering.
By using this mode, mlx4 cards use a fast limited table based steering,
in order to enable fast steering of unicast packets to a QP.
In order to implement A0 hybrid steering we allocate resources
from different zones:
(1) General range
(2) Special MAC-assigned QPs [RSS, Raw-Ethernet] each has its own region.
When we create a rss QP or a raw ethernet (A0 steerable and BF ready) QP,
we try hard to allocate the QP from range (2). Otherwise, we try hard not
to allocate from this range. However, when the system is pushed to its
limits and one needs every resource, the allocator uses every region it can.
Meaning, when we run out of raw-eth qps, the allocator allocates from the
general range (and the special-A0 area is no longer active). If we run out
of RSS qps, the mechanism tries to allocate from the raw-eth QP zone. If that
is also exhausted, the allocator will allocate from the general range
(and the A0 region is no longer active).
Note that if a raw-eth qp is allocated from the general range, it attempts
to allocate the range such that bits 6 and 7 (blueflame bits) in the
QP number are not set.
When the feature is used in SRIOV, the VF has to notify the PF what
kind of QP attributes it needs. In order to do that, along with the
"Eth QP blueflame" bit, we reserve a new "A0 steerable QP". According
to the combination of these bits, the PF tries to allocate a suitable QP.
In order to maintain backward compatibility (with older PFs), the PF
notifies which QP attributes it supports via QUERY_FUNC_CAP command.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When using BF (Blue-Flame), the QPN overrides the VLAN, CV, and SV fields
in the WQE. Thus, BF may only be used for QPNs with bits 6,7 unset.
The current Ethernet driver code reserves a Tx QP range with 256b alignment.
This is wrong because if there are more than 64 Tx QPs in use,
QPNs >= base + 65 will have bits 6/7 set.
This problem is not specific for the Ethernet driver, any entity that
tries to reserve more than 64 BF-enabled QPs should fail. Also, using
ranges is not necessary here and is wasteful.
The new mechanism introduced here will support reservation for
"Eth QPs eligible for BF" for all drivers: bare-metal, multi-PF, and VFs
(when hypervisors support WC in VMs). The flow we use is:
1. In mlx4_en, allocate Tx QPs one by one instead of a range allocation,
and request "BF enabled QPs" if BF is supported for the function
2. In the ALLOC_RES FW command, change param1 to:
a. param1[23:0] - number of QPs
b. param1[31-24] - flags controlling QPs reservation
Bit 31 refers to Eth blueflame supported QPs. Those QPs must have
bits 6 and 7 unset in order to be used in Ethernet.
Bits 24-30 of the flags are currently reserved.
When a function tries to allocate a QP, it states the required attributes
for this QP. Those attributes are considered "best-effort". If an attribute,
such as Ethernet BF enabled QP, is a must-have attribute, the function has
to check that attribute is supported before trying to do the allocation.
In a lower layer of the code, mlx4_qp_reserve_range masks out the bits
which are unsupported. If SRIOV is used, the PF validates those attributes
and masks out unsupported attributes as well. In order to notify VFs which
attributes are supported, the VF uses QUERY_FUNC_CAP command. This command's
mailbox is filled by the PF, which notifies which QP allocation attributes
it supports.
Signed-off-by: Eugenia Emantayev <eugenia@mellanox.co.il>
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull infiniband/rdma fixes from Roland Dreier:
"Last late set of InfiniBand/RDMA fixes for 3.17:
- fixes for the new memory region re-registration support
- iSER initiator error path fixes
- grab bag of small fixes for the qib and ocrdma hardware drivers
- larger set of fixes for mlx4, especially in RoCE mode"
* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (26 commits)
IB/mlx4: Fix VF mac handling in RoCE
IB/mlx4: Do not allow APM under RoCE
IB/mlx4: Don't update QP1 in native mode
IB/mlx4: Avoid accessing netdevice when building RoCE qp1 header
mlx4: Fix mlx4 reg/unreg mac to work properly with 0-mac addresses
IB/core: When marshaling uverbs path, clear unused fields
IB/mlx4: Avoid executing gid task when device is being removed
IB/mlx4: Fix lockdep splat for the iboe lock
IB/mlx4: Get upper dev addresses as RoCE GIDs when port comes up
IB/mlx4: Reorder steps in RoCE GID table initialization
IB/mlx4: Don't duplicate the default RoCE GID
IB/mlx4: Avoid null pointer dereference in mlx4_ib_scan_netdevs()
IB/iser: Bump version to 1.4.1
IB/iser: Allow bind only when connection state is UP
IB/iser: Fix RX/TX CQ resource leak on error flow
RDMA/ocrdma: Use right macro in query AH
RDMA/ocrdma: Resolve L2 address when creating user AH
mlx4: Correct error flows in rereg_mr
IB/qib: Correct reference counting in debugfs qp_stats
IPoIB: Remove unnecessary port query
...
We had several problems here. First, a race condition on QP1 mac
handling between mlx4_ib_update_qps and mlx4_ib_modify_qp, which is
fixed by taking the qp mutex in mlx4_ib_update_qps.
Also, qp->pri.smac_port was not updated in mlx4_ib_update_qps.
Last, in __mlx4_ib_modify_qp we did not properly handle the case where
the mac is zero, but port is non-zero.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Automatic Path Migration is not supported under RoCE. Therefore,
return a "not-supported" error if the caller attempts to set an
alternate path in a QP context.
In addition, if there are no IB ports configured, do not report
APM capability in the device flags returned by mlx4_ib_query_device.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The source MAC is needed in RoCE when building the QP1 header.
Currently, this is obtained from the source net device. However, the net
device may not yet exist, or can be destroyed in parallel to this QP1 send
operation (e.g through the VPI port change flow) so accessing it may cause
a kernel crash.
To fix this, we maintain a source MAC cache per port for the net device in
struct mlx4_ib_roce. This cached MAC is initialized to be the default MAC
address obtained during HCA initialization via QUERY_PORT. This cached MAC
is updated via the netdev event notifier handler.
Since the cached MAC is held in an atomic64 object, we do not need locking
when accessing it.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
We wrongly tested QP context bits without BE conversion
as was spotted by sparse...
drivers/infiniband/hw/mlx4/qp.c:1685:38: sparse: restricted __be32 degrades to integer
Fix that!
Fixes: d2fce8a ('mlx4: Set user-space raw Ethernet QPs to properly handle VXLAN traffic')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Raw Ethernet QPs opened from user-space lack the proper setup to
recieve/handle VXLAN traffic when VXLAN offloads are enabled.
Fix that by adding a tunnel steering rule on top of the normal unicast
steering rule and set the tunnel_type field in the QP context.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull main InfiniBand/RDMA updates from Roland Dreier:
- add iWARP port mapper to avoid conflicts between RDMA and normal
stack TCP connections.
- fixes for i386 / x86-64 structure padding differences (ABI
compatibility for 32-on-64) from Yann Droneaud.
- a pile of SRP initiator fixes from Bart Van Assche.
- fixes for a writeback / memory allocation deadlock with NFS over
IPoIB connected mode from Jiri Kosina.
- the usual fixes and cleanups to mlx4, mlx5, cxgb4 and other low-level
drivers.
* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (61 commits)
RDMA/cxgb4: Add support for iWARP Port Mapper user space service
RDMA/nes: Add support for iWARP Port Mapper user space service
RDMA/core: Add support for iWARP Port Mapper user space service
IB/mlx4: Fix gfp passing in create_qp_common()
IB/umad: Fix use-after-free on close
IB/core: Fix kobject leak on device register error flow
RDMA/cxgb4: add missing padding at end of struct c4iw_alloc_ucontext_resp
mlx4_core: Fix GFP flags parameters to be gfp_t
IB/core: Fix port kobject deletion during error flow
IB/core: Remove unneeded kobject_get/put calls
IB/core: Fix sparse warnings about redeclared functions
IB/mad: Fix sparse warning about gfp_t use
IB/mlx4: Implement IB_QP_CREATE_USE_GFP_NOIO
IB: Add a QP creation flag to use GFP_NOIO allocations
IB: Return error for unsupported QP creation flags
IB: Allow build of hw/ and ulp/ subdirectories independently
mlx4_core: Move handling of MLX4_QP_ST_MLX to proper switch statement
RDMA/cxgb4: Add missing padding at end of struct c4iw_create_cq_resp
IB/srp: Avoid problems if a header uses pr_fmt
IB/umad: Fix error handling
...
There are two kzalloc() calls which were not converted to use value of
gfp passed to create_qp_common() instead of using hardcoded GFP_KERNEL
in 40f2287bd5 ("IB/mlx4: Implement IB_QP_CREATE_USE_GFP_NOIO"). Fix
this by passing gfp value down properly.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Modify the various routines used to allocate memory resources which
serve QPs in mlx4 to get an input GFP directive. Have the Ethernet
driver to use GFP_KERNEL in it's QP allocations as done prior to this
commit, and the IB driver to use GFP_NOIO when the IB verbs
IB_QP_CREATE_USE_GFP_NOIO QP creation flag is provided.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Eran Ben Elisha
Moni Shoua
Doug Ledford
Sebastian Ott
Erez Shitrit
Linus Torvalds
Majd Dibbiny
Yishai Hadas
Matan Barak
Eugenia Emantayev
Jack Morgenstein
Or Gerlitz
Roland Dreier
Jiri Kosina