Leon Romanovsky [Tue, 20 Jun 2017 04:55:53 +0000 (07:55 +0300)]
RDMA/netlink: Add netlink device definitions to UAPI
Introduce new defines to rdma_netlink.h, so the RDMA configuration tool
will be able to communicate with RDMA subsystem by using the shared defines.
The addition of new client (NLDEV) revealed the fact that we exposed by
mistake the RDMA_NL_I40IW define which is not backed by any RDMA netlink
by now and it won't be exposed in the future too. So this patch reuses
the value and deletes the old defines.
The NLDEV operates with objects. The struct ib_device has two straightforward
objects: device itself and ports of that device.
This brings us to propose the following commands to work on those objects:
* RDMA_NLDEV_CMD_{GET,SET,NEW,DEL} - works on ib_device itself
* RDMA_NLDEV_CMD_PORT_{GET,SET,NEW,DEL} - works on ports of specific ib_device
Those commands receive/return the device index (RDMA_NLDEV_ATTR_DEV_INDEX)
and port index (RDMA_NLDEV_ATTR_PORT_INDEX). For device object accesses,
the RDMA_NLDEV_ATTR_PORT_INDEX will return the maximum number of ports
for specific ib_device and for port access the actual port index.
The port index starts from 1 to follow RDMA/core internal semantics and
the sysfs exposed knobs.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Leon Romanovsky [Thu, 15 Jun 2017 09:46:33 +0000 (12:46 +0300)]
RDMA/netlink: Add and implement doit netlink callback
The .doit callback is used by netlink core to differentiate
between get and set operations. Common convention is to use
that call for command operations like (SET, ADD, e.t.c.) and/or
access without NLF_M_DUMP flag.
This commit adds proper declaration and implementation
to RDMA netlink.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Leon Romanovsky [Sun, 18 Jun 2017 11:39:59 +0000 (14:39 +0300)]
RDMA/core: Add and expose static device index
This patch adds static device index in similar fashion to
already available in netdev world (struct net->ifindex).
In downstream patches, the RDMA nelink will use this idx-to-ib_device
conversion, so as part of this commit, we are exposing the translation
function to be visible for IB/core users.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Leon Romanovsky [Mon, 19 Jun 2017 11:04:56 +0000 (14:04 +0300)]
RDMA/core: Add iterator over ib_devices
The coming nldev needs to iterate over all IB devices in the system
and in order to not expose the ib_devices list outside the devices.c,
it is necessary to provide function iterator.
Current version is written explicitly for nldev callback to avoid
over-engineering at this stage, but it can be easily extended for
other types.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Leon Romanovsky [Mon, 19 Jun 2017 15:23:45 +0000 (18:23 +0300)]
RDMA/netlink: Rename netlink callback struct
The RDMA netlink client infrastructure was removed and made obsolete.
The old infrastructure defined struct ibnl_client_cbs. Now that all
uses of this have been updated to the new infrastructure, rename the
struct to be compliant with the current stack naming standards:
struct rdma_nl_cbs.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Leon Romanovsky [Mon, 12 Jun 2017 13:00:19 +0000 (16:00 +0300)]
RDMA/netlink: Add flag to consolidate common handling
Add ability to provide flags to control RDMA netlink callbacks
and convert addr.c and sa_query.c to be first users of such
infrastructure. It allows to move their CAP_NET_ADMIN checks
into netlink core.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com>
RDMA netlink has a complicated infrastructure for dynamically
registering and de-registering netlink clients to the NETLINK_RDMA
group. The complicated portion of this code is not widely used because
2 of the 3 current clients are statically compiled together with
netlink.c. The infrastructure, therefore, is deemed overkill.
Refactor the code to eliminate the dynamically added clients. Now all
clients are pre-registered in a client array at compile time, and at run
time they merely check-in with the infrastructure to pass their callback
table for inclusion in the pre-sized client array.
This also allows for future cleanups and removal of unneeded code in the
iwcm* netlink handler.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Chien Tin Tung <chien.tin.tung@intel.com>
Ismail, Mustafa [Wed, 28 Jun 2017 14:02:45 +0000 (09:02 -0500)]
RDMA/core: Add wait/retry version of ibnl_unicast
Add a wait/retry version of ibnl_unicast, ibnl_unicast_wait,
and modify ibnl_unicast to not wait/retry. This eliminates
the undesirable wait for future users of ibnl_unicast.
Change Portmapper calls originating from kernel to user-space
to use ibnl_unicast_wait and take advantage of the wait/retry
logic in netlink_unicast.
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com> Signed-off-by: Chien Tin Tung <chien.tin.tung@intel.com> Signed-off-by: Leon Romanovsky <leon@kernel.org>
Always initiate an offline transition request
when a link down occurs. The firmware will
use this request to confirm that the driver
has seen the link down message. A host version
is set to indicate this driver behavior to the
firmware.
Reviewed-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
When link interrupts occur, multiple link down requests
could be queued up when only one is needed. This could get
the hfi1 out of sync with its link partner during LNI.
Only allow one link down request to be queued at any one time.
Reviewed-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Currently, link down interrupts queue link entries
on a workqueue intended for sending events only.
Create a workqueue for queuing link events.
Reviewed-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
The root cause is that the qib/hfi1 post receive call to rvt_lkey_ok()
doesn't interpret the new return value from rvt_lkey_ok() properly
leading to an mr reference count underrun.
Additionally, remove an unused argument in rvt_sge_adjacent()
aw well as an unneeded incr local in rvt_post_one_wr().
Fixes: Commit 14fe13fcd3af ("IB/rdmavt: Compress adjacent SGEs in rvt_lkey_ok()") Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Jan Sokolowski [Sat, 29 Jul 2017 15:43:37 +0000 (08:43 -0700)]
IB/hfi1: Disambiguate corruption and uninitialized error cases
The error messages when checksum validation of the platform
configuration fields populated into the ASIC scratch registers fails are
ambiguous. Disambiguate them.
Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com> Signed-off-by: Easwar Hariharan <easwar.hariharan@intel.com> Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Sat, 29 Jul 2017 15:43:32 +0000 (08:43 -0700)]
IB/hfi1: Only set fd pointer when base context is completely initialized
The allocate_ctxt() function adds the context to the fd data structure.
Since the context is not completely initialized, this can cause confusion
as to whether the context is valid or not.
Move the fd reference from allocate_ctxt() to setup_base_ctxt().
Update the necessary functions to be aware of this move.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Jan Sokolowski [Sat, 29 Jul 2017 15:43:26 +0000 (08:43 -0700)]
IB/hfi1: Do not enable disabled port on cable insert
Fix issue where a disabled port can be enabled by
inserting a cable. The port should be explicitly
enabled instead.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com> Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Alex Estrin [Sat, 29 Jul 2017 15:43:20 +0000 (08:43 -0700)]
IB/hfi1: Harden state transition to Armed and Active
There is a window that allows other threads to read state of
'host_link_state' as a new, before the hardware actual state is set.
This patch closes the window by indicating a new state only after
hardware transition is complete.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Alex Estrin <alex.estrin@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:46:42 +0000 (07:46 -0700)]
IB/hfi1: Split copy_to_user data copy for better security
A copy_to_user() call assumes that two members of a data structure
are sequential. Since this may not always be true, separate the copies
to ensure a safe copy.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Alex Estrin [Mon, 24 Jul 2017 14:46:36 +0000 (07:46 -0700)]
IB/hfi1: Verify port data VLs credits on transition to Armed
There is a window where the FM can read the buffer control table
and decide not to program buffers. When a port goes down, the code
clears the table and if it is not programmed, posted SDMA descriptors
will never complete due to no buffer credits.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Alex Estrin <alex.estrin@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/hfi1: Move saving PCI values to a separate function
During PCIe initialization some registers' values from
PCI config space are saved in order to restore them later
(i.e. after reset). Restoring those value is done by a
function called restore_pci_variables, while saving them
is put directly into function hfi1_pcie_ddinit.
Move saving values to a separate function in the image
of restoring functionality.
Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Bartlomiej Dudek <bartlomiej.dudek@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/hfi1: Fix initialization failure for debug firmware
Loading debug signed firmware fails if started immediately after
failed attempt to load production firmware. A short delay is
required so add about a 100us delay after RSA check failure.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jakub Byczkowski <jakub.byczkowski@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Jan Sokolowski [Mon, 24 Jul 2017 14:46:18 +0000 (07:46 -0700)]
IB/hfi1: Fix code consistency for if/else blocks in chip.c
Code structure is not consistent for if/else blocks and break
instructions in set_link_state for case HLS_UP_INIT. Physical
state uses break in case of an error and if/else blocks for
logical use cases. These blocks should be implemented consistently.
Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com> Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:46:12 +0000 (07:46 -0700)]
IB/hfi1: Send MAD traps until repressed
A trap should be sent to the FM until the FM sends a repress message.
This is in line with the IBTA 13.4.9.
Add the ability to resend traps until a repress message is received.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Michael N. Henry <michael.n.henry@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:46:06 +0000 (07:46 -0700)]
IB/hfi1: Pass the context pointer rather than the index
The hfi1_rcvctrl() function receives an index which it then converts
to an rcd. Since most functions have the rcd, use that instead.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:46:01 +0000 (07:46 -0700)]
IB/hfi1: Use context pointer rather than context index
The hfi1_<set|clear>_ctxt_<j|p>key functions take a context index and
look up the context based on that index.
Since the context index is being retrieved from the context, this
doesn't seem optimal.
Pass the context pointer for use, rather than the context index.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:45:55 +0000 (07:45 -0700)]
IB/hfi1: Size rcd array index correctly and consistently
The array index for the rcd array is sized several different ways
throughout the code.
Use the user interface size (u16) as the standard size and update the
necessary code to reflect this.
u16 is large enough for the largest amount of supported contexts.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:45:49 +0000 (07:45 -0700)]
IB/hfi1: Remove unused user context data members
Several data members of the user context have become unused over time.
Cleaning them up.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Ira Weiny <ira.weiny@intel.com> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Michael J. Ruhl [Mon, 24 Jul 2017 14:45:43 +0000 (07:45 -0700)]
IB/hfi1: Assign context does not clean up file descriptor correctly on error
In the error path for context allocation, the file descriptor pointer
should not point to a context when an error occurs.
Clean up the appropriate references on error.
Fixes: Commit 62239fc6e5545b2e59f83dfbc5db231a81f37a45 ("IB/hfi1: Clean up on context initialization failure") Reviewed-by: Ira Weiny <ira.weiny@intel.com> Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Kaike Wan [Mon, 24 Jul 2017 14:45:37 +0000 (07:45 -0700)]
IB/hfi1: Serve the most starved iowait entry first
When an egress resource(SDMA descriptors, pio credits) is not available,
a sending thread will be put on the resource's wait queue. When the
resource becomes available again, up to a fixed number of sending threads
can be awakened sequentially and removed from the wait queue, depending
on the number of waiting threads and the number of free resources. Since
each awakened sending thread will send as many packets as possible, it
is highly likely that the first sending thread will consume all the
egress resources. Subsequently, it will be put back to the end of the wait
queue. Depending on the timing when the later sending threads wake up,
they may not be able to send any packet and be again put back to the end
of the wait queue sequentially, right behind the first sending thread.
This starvation cycle continues until some sending threads exceed their
retry limit and consequently fail.
This patch fixes the issue by two simple approaches:
(1) Any starved sending thread will be put to the head of the wait queue
while a served sending thread will be put to the tail;
(2) The most starved sending thread will be served first.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Kaike Wan <kaike.wan@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
The ioremap_wc() clearly is not returning a write combining mapping due
to an overlap where the RcvArray is mapped in a uncached mapping prior
to creating the proposed write combining mapping.
The patch replaces the single base register for uncached CSRs that
used to overlap the RcvArray with two mappings. One, kregbase1, from the
bar0 up to the RcvArray and another, kregbase2, from the end of the
RcvArray to the pio send buffer space. A new dd field, base2_start,
is used to convert the zero-based offset in the CSR routines to the
correct kregbase1/kregbase2 mapping. A single direct write of the
RcvArray CSRs is replaced with hfi1_put_tid() to insure correct access
using the new disjoint mapping.
Additionally, the kregend field is deleted since it is only ever written.
patdebug now shows the RcvArray as write combining:
[ 35.688990] x86/PAT: reserve_memtype added [mem 0x91200000-0x9127ffff],
track write-combining, req write-combining, ret write-combining
To insulate from any potential issues with write combining, all
writeq are now flushed in hfi1_put_tid() and rcv_array_wc_fill().
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com> Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Bartlomiej Dudek [Fri, 30 Jun 2017 20:14:40 +0000 (13:14 -0700)]
IB/hfi1: Check return values from PCI config API calls
Ensure that return values from kernel PCI config access
API calls in HFI driver are checked and react properly if
they are not expected (i.e. not successful).
Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com> Signed-off-by: Bartlomiej Dudek <bartlomiej.dudek@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/infiniband/hw/hns/hns_roce_eq.c:477:8: error: unknown type name 'irqreturn_t'
static irqreturn_t hns_roce_msi_x_interrupt(int irq, void *eq_ptr)
^~~~~~~~~~~
drivers/infiniband/hw/hns/hns_roce_eq.c: In function 'hns_roce_msi_x_interrupt':
drivers/infiniband/hw/hns/hns_roce_eq.c:485:9: error: implicit declaration of function 'IRQ_RETVAL'; did you mean 'BPF_RVAL'? [-Werror=implicit-function-declaration]
return IRQ_RETVAL(int_work);
I have bisected this to a seemingly unrelated change that happened
to remove some indirect header inclusions. Simply including the
required header explicitly fixes the build failure.
The 0day build system flags implicit includes as errors. A patch from
Matan Barak to allow hns_roce, an aarch64 specific RDMA driver, to be
built on other arches, but it resulted in build regressions. The
problem is that hns_roce_device.h needs a definition for __raw_writeq
but did not have an include to provide it. Add <linux/io.h> as an
include to resolve the issue.
Matan Barak [Thu, 8 Jun 2017 14:23:50 +0000 (17:23 +0300)]
IB/hns: Support compile test for hns RoCE driver
Compiling the hns RoCE driver requires ARM architecture.
In order to simplify development of IB/core, support
compile test. Add the necessary includes for that too.
The initial patch for changing the stack to use RoCEv2 GIDs by default
set the CMA_PREFERRED_ROCE_GID_TYPE to an incorrect value. Instead of
an absolute value, we needed to set the right bit in a bitmask. Correct
the default setting so we use RoCEv2 by default.
Fixes: 63a5f483af0e (IB/cma: Set default gid type to RoCEv2) Signed-off-by: Doug Ledford <dledford@redhat.com>
Amrani, Ram [Mon, 26 Jun 2017 16:05:06 +0000 (19:05 +0300)]
RDMA/qedr: notify user application of supported WIDs
The number of supported WIDs, if they are supported at all, can be
limited due to resources. Notifying the user space application the
number of available WIDs allows it to utilize them correctly.
Amrani, Ram [Mon, 26 Jun 2017 16:05:05 +0000 (19:05 +0300)]
RDMA/qedr: notify user application if DPM is supported
Direct Packet Mode support may be disabled, e.g, due to limited
resources. Notifying the user application prevents wasting cycles
on attempting to send these kind of packets.
Guy Levi [Tue, 4 Jul 2017 13:24:26 +0000 (16:24 +0300)]
IB/mlx4: Add support for RSS QP
Add support to work with a RSS QP by using an indirection table object
upon QP creation. Other related QP verbs (e.g. modify/destroy/query) were
updated as well for that QP mode.
Notes:
- The RX hash properties are supplied as driver private data.
- The RSS QP port is used on the associated WQs in its indirection
table. Applying different ports during WQ life time is not allowed.
- The expected RSS QP flow is: create, modify(RST->INIT),
modify(RST->RTR), destroy.
Signed-off-by: Guy Levi <guyle@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Guy Levi [Tue, 4 Jul 2017 13:24:25 +0000 (16:24 +0300)]
IB/mlx4: Add support for WQ indirection table related verbs
To enable RSS functionality the IB indirection table object (i.e.
ib_rwq_ind_table) should be used.
This patch implements the related verbs as of create and destroy an
indirection table.
In downstream patches the indirection table will be used as part of RSS
QP creation.
Signed-off-by: Guy Levi <guyle@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Guy Levi [Tue, 4 Jul 2017 13:24:24 +0000 (16:24 +0300)]
IB/mlx4: Add support for WQ related verbs
Support create/modify/destroy WQ related verbs.
The base IB object to enable RSS functionality is a WQ (i.e. ib_wq).
This patch implements the related WQ verbs as of create, modify and
destroy.
In downstream patches the WQ will be used as part of an indirection
table (i.e. ib_rwq_ind_table) to enable RSS QP creation.
Notes:
ConnectX-3 hardware requires consecutive WQNs list as receive descriptor
queues for the RSS QP. Hence, the driver manages consecutive ranges lists
per context which the user must respect.
Destroying the WQ does not return its WQN back to its range for
reusing. However, destroying all WQs from the same range releases the
range and in turn releases its WQNs for reusing.
Since the WQ object is not a natural object in the hardware, the driver
implements the WQ by the hardware QP.
As such, the WQ inherits its port from its RSS QP parent upon its
RST->INIT transition and by that time its state is applied to the
hardware.
Signed-off-by: Guy Levi <guyle@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Moshe Shemesh [Wed, 21 Jun 2017 06:29:36 +0000 (09:29 +0300)]
(IB, net)/mlx4: Add resource utilization support
Adding visibility of resource usage of QPs, CQs and counters used by
virtual functions. This feature will be used to give the PF administrator
more data while debugging VF status. Usage info was added to ALLOC_RES
command, to notify the PF if the resource which is being reserved or
allocated for the VF will be used by kernel driver or by user verbs.
Updated reservation and allocation functions of QP, CQ and counter with
additional usage parameter.
Maor Gottlieb [Wed, 21 Jun 2017 06:26:28 +0000 (09:26 +0300)]
IB/mlx4: Add inline-receive support
When inline-receive is enabled, the HCA may write received
data into the receive WQE.
Inline-receive is enabled by setting its matching bit in
the QP context and each single-packet message with payload
not exceeding the receive WQE size will be delivered to
the WQE.
The completion report will indicate that the payload was placed to the WQE.
It includes:
1) Return maximum supported size of inline-receive by the hardware
in query_device vendor's data part.
2) Enable the feature when requested by the vendor data input.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Parav Pandit [Mon, 19 Jun 2017 04:19:37 +0000 (07:19 +0300)]
IB/mlx5: Expose extended error counters
This patch adds below requester and responder side error counters,
which will be exposed by hardware counters interface and are supported
as part of query Q counters command extension.
+---------------------------+-------------------------------------+
| Name | Description |
|---------------------------+-------------------------------------|
|resp_local_length_error | Number of times responder detected |
| | local length errors |
|---------------------------+-------------------------------------|
|resp_cqe_error | Number of CQEs completed with error |
| | at responder |
|---------------------------+-------------------------------------|
|req_cqe_error | Number of CQEs completed with error |
| | at requester |
|---------------------------+-------------------------------------|
|req_remote_invalid_request | Number of times requester detected |
| | remote invalid request error |
|---------------------------+-------------------------------------|
|req_remote_access_error | Number of times requester detected |
| | remote access error |
|---------------------------+-------------------------------------|
|resp_remote_access_error | Number of times responder detected |
| | remote access error |
|---------------------------+-------------------------------------|
|resp_cqe_flush_error | Number of CQEs completed with |
| | flushed with error at responder |
|---------------------------+-------------------------------------|
|req_cqe_flush_error | Number of CQEs completed with |
| | flushed with error at requester |
+---------------------------+-------------------------------------+
Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Eli Cohen <eli@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Leon Romanovsky [Mon, 12 Jun 2017 07:36:16 +0000 (10:36 +0300)]
IB/mlx5: Fix existence check for extended address vector
The extended address vector is the highest bit in be32 variable,
but it was compared with the lowest. This patch fixes the endianness
of that check and removes already declared define.
Majd Dibbiny [Mon, 12 Jun 2017 07:36:15 +0000 (10:36 +0300)]
IB/mlx5: Fix cached MR allocation flow
When we have a miss in one order of the mkey cache, we try to get
an mkey from a higher order.
We still need to check that the higher order can be used with UMR
before using it. Otherwise, we will get an mkey with 0 entries and
the post send operation that is used to fill it will complete with
the following error:
Yishai Hadas [Thu, 8 Jun 2017 13:15:09 +0000 (16:15 +0300)]
IB/mlx5: Add multicast flow steering support for underlay QP
In order to add multicast flow steering support, there is need
to block the attaching of mcg flow for underlay QP, recognize
multicast IB_FLOW_SPEC_IPV4 based on its IP and enable
creating/destroying flow for IB layer.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Maor Gottlieb <maorg@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Yishai Hadas [Thu, 8 Jun 2017 13:15:08 +0000 (16:15 +0300)]
IB/mlx5: Add support for QP with a given source QPN
Allow user space applications to accelerate send and receive
traffic which is typically handled by IPoIB ULP by creating
a UD QP with a given source QPN of the IPoIB UD QP.
UD QP with a given source QPN should basically be similar to
RAW QP from point of view of its created resources.
However,
- Its TIS should point to the source QPN.
- Modify can be done only on its state as the transport attributes
are managed by its source QP.
This patch manages below:
- Creating/destroying/modifying UD QP with a given source QPN.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Maor Gottlieb <maorg@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Yishai Hadas [Thu, 8 Jun 2017 13:15:07 +0000 (16:15 +0300)]
IB/uverbs: Enable QP creation with a given source QP number
Enable QP creation with a given source QP number, the created QP will
use the source QPN as its wire QP number.
To create such a QP, root privileges (i.e. CAP_NET_RAW) are required
from the user application.
This comes as a pre-patch for downstream patches in this series to
allow user space applications to accelerate traffic which is typically
handled by IPoIB ULP.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Maor Gottlieb <maorg@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Yishai Hadas [Thu, 8 Jun 2017 13:15:06 +0000 (16:15 +0300)]
IB/core: Enable QP creation with a given source QP number
Enable QP creation with a given source QP number.
The created QP will use the source QPN as its wire QP number.
This comes as a pre-patch for downstream patches in this series to
allow user space applications to accelerate traffic which is typically
handled by IPoIB ULP.
Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Maor Gottlieb <maorg@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Noa Osherovich [Mon, 12 Jun 2017 08:14:04 +0000 (11:14 +0300)]
IB/core: Add support for RoCEv2 multicast
When creating address handle from multicast GID, set MAC according to
the appropriate formula instead of searching for it in the GID table:
- For IPv4 multicast GID use ip_eth_mc_map().
- For IPv6 multicast GID use ipv6_eth_mc_map().
Signed-off-by: Noa Osherovich <noaos@mellanox.com> Reviewed-by: Moni Shoua <monis@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Noa Osherovich [Mon, 12 Jun 2017 08:14:03 +0000 (11:14 +0300)]
IB/core: Set RoCEv2 MGID according to spec
RoCEv2 Annex states that for RoCEv2 over IPv4, the corresponding
IPv4 address is encoded into the GID according to the following rule:
GID= :ffff:<IPv4 address>
Remove the 0xff0e prefix for RoCEv2 packets with IPv4 and leave it
zeroed and change rdma_is_multicast_addr() to consider the new logic.
Signed-off-by: Noa Osherovich <noaos@mellanox.com> Reviewed-by: Moni Shoua <monis@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Noa Osherovich [Mon, 12 Jun 2017 08:14:02 +0000 (11:14 +0300)]
IB/core: Fix the validations of a multicast LID in attach or detach operations
RoCE Annex (A16.9.10/11) declares that during attach (detach) QP to a
multicast group, if the QP is associated with a RoCE port, the
multicast group MLID is unused and is ignored.
During attach or detach multicast, when the QP is associated with a
port, it is enough to check the port's link layer and validate the
LID only if it is Infiniband. Otherwise, avoid validating the
multicast LID.
Fixes: 8561eae60ff9 ("IB/core: For multicast functions, verify that LIDs are multicast LIDs") Signed-off-by: Noa Osherovich <noaos@mellanox.com> Reviewed-by: Moni Shoua <monis@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Maor Gottlieb [Tue, 30 May 2017 07:29:13 +0000 (10:29 +0300)]
IB/mlx5: Add support to dropless RQ
RQs that were configured for "delay drop" will prevent packet drops
when their WQEs are depleted.
Marking an RQ to be drop-less is done by setting delay_drop_en in RQ
context using CREATE_RQ command.
Since this feature is globally activated/deactivated by using the
SET_DELAY_DROP command on all the marked RQs, we activated/deactivated
it according to the number of RQs with 'delay_drop' enabled.
When timeout is expired, then the feature is deactivated. Therefore
the driver handles the delay drop timeout event and reactivate it.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Maor Gottlieb [Tue, 30 May 2017 07:29:12 +0000 (10:29 +0300)]
net/mlx5: Introduce general notification event
When delay drop timeout is expired, the firmware raises
general notification event of DELAY_DROP_TIMEOUT subtype.
In addition the feature is disable so the driver have to
reactivate the timeout.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Maor Gottlieb [Tue, 30 May 2017 07:29:11 +0000 (10:29 +0300)]
net/mlx5: Introduce set delay drop command
Add support to SET_DELAY_DROP command.
This command will be used in downstream patches for delay packet drop.
The timeout value should be indicated by delay_drop_timeout field.
Packet processing will be delayed till timeout value passed or until
more WQEs are posted.
Setting this value to 0 disables the feature.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Maor Gottlieb [Tue, 30 May 2017 07:29:10 +0000 (10:29 +0300)]
IB/core: Introduce delay drop for a WQ
Work queue which is created with IB_WQ_FLAGS_DELAY_DROP won't
cause packet drops when there aren't receive WQEs, but will wait until
posting of receive WQEs or for some period of time that the device
was configured with.
It includes:
* Add a new creation flag to enable delay drop functionality in a WQ.
* A new capability was introduced - IB_RAW_PACKET_CAP_DELAY_DROP, which
is the device's ability to delay packet drops when there aren't receive
WQEs.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Bodong Wang [Tue, 30 May 2017 07:18:24 +0000 (10:18 +0300)]
IB/mlx5: Restore IB guid/policy for virtual functions
When a user sets port_guid, node_guid or policy of an IB virtual
function, save this information in "struct mlx5_vf_context".
This information will be restored later when pci_resume is called.
To make sure this works, one can use aer-inject to generate PCI
errors on mlx5 devices and verify if relevant fields are restored
after PCI resume.
Signed-off-by: Bodong Wang <bodong@mellanox.com> Reviewed-by: Eli Cohen <eli@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Parav Pandit [Tue, 30 May 2017 07:05:15 +0000 (10:05 +0300)]
IB/mlx5: Add debug control parameters for congestion control
This patch adds debug control parameters for congestion control which
can be read or written through debugfs. They are for reaction point and
notification point nodes.
These control parameters are as below:
+------------------------------+-----------------------------------------+
| Name | Description |
|------------------------------+-----------------------------------------|
|rp_clamp_tgt_rate | When set target rate is updated to |
| | current rate |
|------------------------------+-----------------------------------------|
|rp_clamp_tgt_rate_ati | When set update target rate based on |
| | timer as well |
|------------------------------+-----------------------------------------|
|rp_time_reset | time between rate increase if no |
| | CNP is received unit in usec |
|------------------------------+-----------------------------------------|
|rp_byte_reset | Number of bytes between rate inease if |
| | no CNP is received |
|------------------------------+-----------------------------------------|
|rp_threshold | Threshold for reaction point rate |
| | control |
|------------------------------+-----------------------------------------|
|rp_ai_rate | Rate for target rate, unit in Mbps |
|------------------------------+-----------------------------------------|
|rp_hai_rate | Rate for hyper increase state |
| | unit in Mbps |
|------------------------------+-----------------------------------------|
|rp_min_dec_fac | Minimum factor by which the current |
| | transmit rate can be changed when |
| | processing a CNP, unit is percerntage |
|------------------------------+-----------------------------------------|
|rp_min_rate | Minimum value for rate limit, |
| | unit in Mbps |
|------------------------------+-----------------------------------------|
|rp_rate_to_set_on_first_cnp | Rate that is set when first CNP is |
| | received, unit is Mbps |
|------------------------------+-----------------------------------------|
|rp_dce_tcp_g | Used to calculate alpha |
|------------------------------+-----------------------------------------|
|rp_dce_tcp_rtt | Time between updates of alpha value, |
| | unit is usec |
|------------------------------+-----------------------------------------|
|rp_rate_reduce_monitor_period | Minimum time between consecutive rate |
| | reductions |
|------------------------------+-----------------------------------------|
|rp_initial_alpha_value | Initial value of alpha |
|------------------------------+-----------------------------------------|
|rp_gd | When CNP is received, flow rate is |
| | reduced based on gd, rp_gd is given as |
| | log2(rp_gd) |
|------------------------------+-----------------------------------------|
|np_cnp_dscp | dscp code point for generated cnp |
|------------------------------+-----------------------------------------|
|np_cnp_prio_mode | 802.1p priority for generated cnp |
|------------------------------+-----------------------------------------|
|np_cnp_prio | cnp priority mode |
+------------------------------+-----------------------------------------+
Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Eli Cohen <eli@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Moni Shoua [Tue, 30 May 2017 06:56:05 +0000 (09:56 +0300)]
IB/mlx5: Change logic for dispatching IB events for port state
The old logic ignored link state. This led to missing IB events like
when link goes down on the switch while admin state is up or to redundant
events like when admin state goes up while link is down.
To fix that, probe the port state on NETDEV events and compare to last
known state to decide if IB events needs to be dispatched.
FIxes: 5ec8c83e3ad3 ("IB/mlx5: Port events in RoCE now rely on netdev events") Signed-off-by: Moni Shoua <monis@mellanox.com> Reviewed-by: Noa Osherovich <noaos@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Huy Nguyen [Tue, 30 May 2017 06:42:54 +0000 (09:42 +0300)]
IB/mlx5: Add raw ethernet local loopback support
Currently, unicast/multicast loopback raw ethernet
(non-RDMA) packets are sent back to the vport.
A unicast loopback packet is the packet with destination
MAC address the same as the source MAC address.
For multicast, the destination MAC address is in the
vport's multicast filter list.
Moreover, the local loopback is not needed if
there is one or none user space context.
After this patch, the raw ethernet unicast and multicast
local loopback are disabled by default. When there is more
than one user space context, the local loopback is enabled.
Note that when local loopback is disabled, raw ethernet
packets are not looped back to the vport and are forwarded
to the next routing level (eswitch, or multihost switch,
or out to the wire depending on the configuration).
Signed-off-by: Huy Nguyen <huyn@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Selvin Xavier [Thu, 29 Jun 2017 19:28:15 +0000 (12:28 -0700)]
RDMA/bnxt_re: Allow posting when QPs are in error
This patch allows driver to post send and receive
requests on QPs which are in error state.
Instead of flushing the QP in the context of polling
error CQEs, the QPs will be added to a flush list
maintained per CQ. QP state is moved to error.
QP is added to flush list if the user moves it
to error state using modify_qp also. After polling the HW
CQ in poll_cq routine, this flush list is traversed
and driver completes work requests on each QP in the flush
list, till the budget expires. The QP is moved out of
flush list during QP destroy or during modify_QP to RESET.
When ULPs post Work Requests while QP is in error state,
driver will store the ULP data and then increment the
QP producer s/w index, without ringing doorbell. It then
schedules a worker to invoke the CQ handler since the
interrupts wont be generated from the HW for this request.
Kalesh AP [Thu, 29 Jun 2017 19:28:10 +0000 (12:28 -0700)]
RDMA/bnxt_re: Add vlan tag for untagged RoCE traffic when PFC is configured
Current implementation does not program vlan header insertion
in RoCE packet if no vlan is configured. Firmware does not add
prority when there is no vlan tag in the packet. Modify the code
to insert vlan header when PFC is enabled on the interface.
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com> Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Leon Romanovsky [Mon, 26 Jun 2017 05:58:22 +0000 (08:58 +0300)]
RDMA: Remove useless MODULE_VERSION
All modules in drivers/infiniband defined and used MODULE_VERSION, which
was pointless because the kernel version describes their state more accurate
then those arbitrary numbers.
Signed-off-by: Leon Romanovsky <leon@kernel.org> Acked-by: Sagi Grimbrg <sagi@grimberg.me> Reviewed-by: Sagi Grimberg <sagi@grimbeg.me> Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Acked-by: Selvin Xavier <selvin.xavier@broadcom.com> Acked-by: Ram Amrani <Ram.Amrani@cavium.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Acked-by: Adit Ranadive <aditr@vmware.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Addresses-Coverity-ID: 1226951 Signed-off-by: Gustavo A. R. Silva <garsilva@embeddedor.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Moni Shoua [Tue, 30 May 2017 06:47:34 +0000 (09:47 +0300)]
IB/cma: Set default gid type to RoCEv2
RoCEv2 is the preferred RDMA protocol for Ethernet link layer because
of its advantages over RoCEv1. For better user experience make it the
default choice for RDMA_CM connections if device/port support it.
Signed-off-by: Moni Shoua <monis@mellanox.com> Reviewed-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>