Use SRCU in a sensible way by removing all MRs in the implicit tree from
the two xarrays (the update operation), then a synchronize, followed by a
normal single threaded teardown.
This is only a little unusual from the normal pattern as there can still
be some work pending in the unbound wq that may also require a workqueue
flush. This is tracked with a single atomic, consolidating the redundant
existing atomics and wait queue.
For understand-ability the entire ODP implicit create/destroy flow now
largely exists in a single pair of functions within odp.c, with a few
support functions for tearing down an unused child.
Link: https://lore.kernel.org/r/20191009160934.3143-13-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Now that the locking is simplified combine pagefault_implicit_mr() with
implicit_mr_get_data() so that we sweep over the idx range only once,
and do the single xlt update at the end, after the child umems are
setup.
This avoids double iteration/xa_loads plus the sketchy failure path if the
xa_load() fails.
Link: https://lore.kernel.org/r/20191009160934.3143-12-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Now that the child MRs are stored in an xarray we can rely on the SRCU
lock to protect the xa_load and use xa_cmpxchg on the slow allocation path
to resolve races with concurrent page fault.
This reduces the scope of the critical section of umem_mutex for implicit
MRs to only cover mlx5_ib_update_xlt, and avoids taking a lock at all if
the child MR is already in the xarray. This makes it consistent with the
normal ODP MR critical section for umem_lock, and the locking approach
used for destroying an unusued implicit child MR.
The MLX5_IB_UPD_XLT_ATOMIC is no longer needed in implicit_get_child_mr()
since it is no longer called with any locks.
Link: https://lore.kernel.org/r/20191009160934.3143-11-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently the child leaves are stored in the shared interval tree and
every lookup for a child must be done under the interval tree rwsem.
This is further complicated by dropping the rwsem during iteration (ie the
odp_lookup(), odp_next() pattern), which requires a very tricky an
difficult to understand locking scheme with SRCU.
Instead reserve the interval tree for the exclusive use of the mmu
notifier related code in umem_odp.c and give each implicit MR a xarray
containing all the child MRs.
Since the size of each child is 1GB of VA, a 1 level xarray will index 64G
of VA, and a 2 level will index 2TB, making xarray a much better
data structure choice than an interval tree.
The locking properties of xarray will be used in the next patches to
rework the implicit ODP locking scheme into something simpler.
At this point, the xarray is locked by the implicit MR's umem_mutex, and
read can also be locked by the odp_srcu.
Link: https://lore.kernel.org/r/20191009160934.3143-10-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The single routine has a very confusing scheme to advance to the next
child MR when working on an implicit parent. This scheme can only be used
when working with an implicit parent and must not be triggered when
working on a normal MR.
Re-arrange things by directly putting all the single-MR stuff into one
function and calling it in a loop for the implicit case. Simplify some of
the error handling in the new pagefault_real_mr() to remove unneeded gotos.
Link: https://lore.kernel.org/r/20191009160934.3143-9-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This makes the routines easier to understand, particularly with respect
the locking requirements of the entire sequence. The implicit_mr_alloc()
had a lot of ifs specializing it to each of the callers, and only a very
small amount of code was actually shared.
Following patches will cause the flow in the two functions to diverge
further.
Link: https://lore.kernel.org/r/20191009160934.3143-7-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This function is intended to loop across each MTT chunk in the implicit
parent that intersects the range [io_virt, io_virt+bnct). But it is has a
confusing construction, so:
- Consistently use imr and odp_imr to refer to the implicit parent
to avoid confusion with the normal mr and odp of the child
- Directly compute the inclusive start/end indexes by shifting. This is
clearer to understand the intent and avoids any errors from unaligned
values of addr
- Iterate directly over the range of MTT indexes, do not make a loop
out of goto
- Follow 'success oriented flow', with goto error unwind
- Directly calculate the range of idx's that need update_xlt
- Ensure that any leaf MR added to the interval tree always results in an
update to the XLT
Link: https://lore.kernel.org/r/20191009160934.3143-6-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
There is a per device xarray storing mkeys that is used to store every
mkey in the system. However, this xarray is now only read by ODP for
certain ODP designated MRs (ODP, implicit ODP, MW, DEVX_INDIRECT).
Create an xarray only for use by ODP, that only contains ODP related
MKeys. This xarray is protected by SRCU and all erases are protected by a
synchronize.
This improves performance:
- All MRs in the odp_mkeys xarray are ODP MRs, so some tests for is_odp()
can be deleted. The xarray will also consume fewer nodes.
- normal MR's are never mixed with ODP MRs in a SRCU data structure so
performance sucking synchronize_srcu() on every MR destruction is not
needed.
- No smp_load_acquire(live) and xa_load() double barrier on read
Due to the SRCU locking scheme care must be taken with the placement of
the xa_store(). Once it completes the MR is immediately visible to other
threads and only through a xa_erase() & synchronize_srcu() cycle could it
be destroyed.
Link: https://lore.kernel.org/r/20191009160934.3143-4-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The locking model for signature is completely different than ODP, do not
share the same xarray that relies on SRCU locking to support ODP.
Simply store the active mlx5_core_sig_ctx's in an xarray when signature
MRs are created and rely on trivial xarray locking to serialize
everything.
The overhead of storing only a handful of SIG related MRs is going to be
much less than an xarray full of every mkey.
Link: https://lore.kernel.org/r/20191009160934.3143-3-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When working with SRCU protected xarrays the xarray itself should be the
SRCU 'update' point. Instead prefetch is using live as the SRCU update
point and this prevents switching the locking design to use the xarray
instead.
To solve this the prefetch must only read from the xarray once, and hold
on to the actual MR pointer for the duration of the async
operation. Incrementing num_pending_prefetch delays destruction of the MR,
so it is suitable.
Prefetch calls directly to the pagefault_mr using the MR pointer and only
does a single xarray lookup.
All the testing if a MR is prefetchable or not is now done only in the
prefetch code and removed from the pagefault critical path.
Link: https://lore.kernel.org/r/20191009160934.3143-2-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Linux 5.4-rc5
For dependencies in the next patches
Conflict resolved by keeping the delete of the unlock.
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This change allows the RDMA stack to use physical resource numbers if they
are passed up from the device. This is accomplished by separating the
concept of the QP number from the QP handle. Previously, the two were the
same, as the QP number was exposed to the guest and also used to reference
a virtual QP in the device backend.
With physical resource numbers exposed, the QP number given to the guest
is the number assigned from the physical HCA's QP, while the QP handle is
still the internal handle used to reference a virtual QP. Regardless of
whether the device is exposing physical ids, the driver will still try to
pick up the QP handle from the backend if possible. The MR keys exposed to
the guest will also be the MR keys created by the physical HCA, instead of
virtual MR keys. The distinction between handle and keys is already
present for MRs so there is no need to do anything special here.
A new version of the create QP response has been added to the device API
to pass up the QP number and handle. The driver will also report these to
userspace in the udata response if userspace supports it or not create the
queuepair if not. I also had to do a refactor of the destroy qp code to
reuse it if we fail to copy to userspace.
Link: https://lore.kernel.org/r/20191028181444.19448-1-aditr@vmware.com
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Re-design of the iWARP CM related objects reference counting and
synchronization methods, to ensure operations are synchronized correctly
and that memory allocated for "ep" is properly released. Also makes sure
QP memory is not released before ep is finished accessing it.
Where as the QP object is created/destroyed by external operations, the ep
is created/destroyed by internal operations and represents the tcp
connection associated with the QP.
QP destruction flow:
- needs to wait for ep establishment to complete (either successfully or
with error)
- needs to wait for ep disconnect to be fully posted to avoid a race
condition of disconnect being called after reset.
- both the operations above don't always happen, so we use atomic flags to
indicate whether the qp destruction flow needs to wait for these
completions or not, if the destroy is called before these operations
began, the flows will check the flags and not execute them ( connect /
disconnect).
We use completion structure for waiting for the completions mentioned
above.
The QP refcnt was modified to kref object. The EP has a kref added to it
to handle additional worker thread accessing it.
Memory Leaks - https://www.spinics.net/lists/linux-rdma/msg83762.html
Concurrency not managed correctly -
https://www.spinics.net/lists/linux-rdma/msg67949.html
Fixes: de0089e692 ("RDMA/qedr: Add iWARP connection management qp related callbacks")
Link: https://lore.kernel.org/r/20191027200451.28187-4-michal.kalderon@marvell.com
Reported-by: Chuck Lever <chuck.lever@oracle.com>
Reported-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently, the error return path when the call to function
dev->dfx->query_cqc_info fails will leak object 'context'. Fix this by
making the error return path via 'err' return return codes rather than
-EMSGSIZE, set ret appropriately for all error return paths and for the
memory leak now return via 'err' rather than just returning without
freeing context.
Link: https://lore.kernel.org/r/20191024131034.19989-1-colin.king@canonical.com
Addresses-Coverity: ("Resource leak")
Fixes: e1c9a0dc29 ("RDMA/hns: Dump detailed driver-specific CQ")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The "ucmd->log_sq_bb_count" variable is a user controlled variable in the
0-255 range. If we shift more than then number of bits in an int then
it's undefined behavior (it shift wraps), and potentially the int could
become negative.
Fixes: 9a4435375c ("IB/hns: Add driver files for hns RoCE driver")
Link: https://lore.kernel.org/r/20190608092514.GC28890@mwanda
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Provide an ODP explicit/implicit type as part of 'rdma -dd resource show
mr' dump.
For example:
$ rdma -dd resource show mr
dev mlx5_0 mrn 1 rkey 0xa99a lkey 0xa99a mrlen 50000000
pdn 9 pid 7372 comm ibv_rc_pingpong drv_odp explicit
For non-ODP MRs, we won't print "drv_odp ..." at all.
Link: https://lore.kernel.org/r/20191016062308.11886-4-leon@kernel.org
Signed-off-by: Erez Alfasi <ereza@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Introduce ODP diagnostic counters and count the following
per MR within IB/mlx5 driver:
1) Page faults:
Total number of faulted pages.
2) Page invalidations:
Total number of pages invalidated by the OS during all
invalidation events. The translations can be no longer
valid due to either non-present pages or mapping changes.
Link: https://lore.kernel.org/r/20191016062308.11886-2-leon@kernel.org
Signed-off-by: Erez Alfasi <ereza@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
From Yamin Friedman:
====================
This series from Yamin implements long standing "TODO" existed in rw.c. It
allows the driver to specify a cut-over point where it is faster to build
a lkey MR rather than do a large SGL for RDMA READ operations.
mlx5 HW gets a notable performane boost by switching to MRs.
====================
Based on the mlx5-next branch from
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux for
dependencies
* branch 'mlx5-rd-sgl': (3 commits)
RDMA/mlx5: Add capability for max sge to get optimized performance
RDMA/rw: Support threshold for registration vs scattering to local pages
net/mlx5: Expose optimal performance scatter entries capability
Allows the IB device to provide a value of maximum scatter gather entries
per RDMA READ.
In certain cases it may be preferable for a device to perform UMR memory
registration rather than have many scatter entries in a single RDMA READ.
This provides a significant performance increase in devices capable of
using different memory registration schemes based on the number of scatter
gather entries. This general capability allows each device vendor to fine
tune when it is better to use memory registration.
Link: https://lore.kernel.org/r/20191007135933.12483-4-leon@kernel.org
Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The parameters npages used to initial mtt of srq->idx_que shouldn't be
same with srq's. And page_shift should be calculated from idx_buf_pg_sz.
This patch fixes above issues and use field named npage and page_shift
in hns_roce_buf instead of two temporary variables to let us use them
anywhere.
Fixes: 18df508c79 ("RDMA/hns: Remove if-else judgment statements for creating srq")
Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Link: https://lore.kernel.org/r/1567566885-23088-3-git-send-email-liweihang@hisilicon.com
Signed-off-by: Doug Ledford <dledford@redhat.com>
pass_accept_req() is using the same skb for handling accept request and
sending accept reply to HW. Here req and rpl structures are pointing to
same skb->data which is over written by INIT_TP_WR() and leads to
accessing corrupt req fields in accept_cr() while checking for ECN flags.
Reordered code in accept_cr() to fetch correct req fields.
Fixes: 92e7ae7172 ("iw_cxgb4: Choose appropriate hw mtu index and ISS for iWARP connections")
Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Link: https://lore.kernel.org/r/20191003104353.11590-1-bharat@chelsio.com
Signed-off-by: Doug Ledford <dledford@redhat.com>
A TID RDMA READ request could be retried under one of the following
conditions:
- The RC retry timer expires;
- A later TID RDMA READ RESP packet is received before the next
expected one.
For the latter, under normal conditions, the PSN in IB space is used
for comparison. More specifically, the IB PSN in the incoming TID RDMA
READ RESP packet is compared with the last IB PSN of a given TID RDMA
READ request to determine if the request should be retried. This is
similar to the retry logic for noraml RDMA READ request.
However, if a TID RDMA READ RESP packet is lost due to congestion,
header suppresion will be disabled and each incoming packet will raise
an interrupt until the hardware flow is reloaded. Under this condition,
each packet KDETH PSN will be checked by software against r_next_psn
and a retry will be requested if the packet KDETH PSN is later than
r_next_psn. Since each TID RDMA READ segment could have up to 64
packets and each TID RDMA READ request could have many segments, we
could make far more retries under such conditions, and thus leading to
RETRY_EXC_ERR status.
This patch fixes the issue by removing the retry when the incoming
packet KDETH PSN is later than r_next_psn. Instead, it resorts to
RC timer and normal IB PSN comparison for any request retry.
Fixes: 9905bf06e8 ("IB/hfi1: Add functions to receive TID RDMA READ response")
Cc: <stable@vger.kernel.org>
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>
Link: https://lore.kernel.org/r/20191004204035.26542.41684.stgit@awfm-01.aw.intel.com
Signed-off-by: Doug Ledford <dledford@redhat.com>
Broadcom's 575xx adapter series has support for SRIOV VFs. Making changes
to enable SRIOV VF support. There are two major area where changes are
done:
- Added new DB location for control-path and data-path DB ring
- New devices do not need to issue the sriov-config slow-path command
thus, skipping to call that firmware command.
For now enabling support for 64 RoCE VFs.
Link: https://lore.kernel.org/r/1570081715-14301-1-git-send-email-devesh.sharma@broadcom.com
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
While MR uses live as the SRCU 'update', the MW case uses the xarray
directly, xa_erase() causes the MW to become inaccessible to the pagefault
thread.
Thus whenever a MW is removed from the xarray we must synchronize_srcu()
before freeing it.
This must be done before freeing the mkey as re-use of the mkey while the
pagefault thread is using the stale mkey is undesirable.
Add the missing synchronizes to MW and DEVX indirect mkey and delete the
bogus protection against double destroy in mlx5_core_destroy_mkey()
Fixes: 534fd7aac5 ("IB/mlx5: Manage indirection mkey upon DEVX flow for ODP")
Fixes: 6aec21f6a8 ("IB/mlx5: Page faults handling infrastructure")
Link: https://lore.kernel.org/r/20191001153821.23621-7-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
live is used to signal to the pagefault thread that the MR is initialized
and ready for use. It should be after the umem is assigned and all other
setup is completed. This prevents races (at least) of the form:
CPU0 CPU1
mlx5_ib_alloc_implicit_mr()
implicit_mr_alloc()
live = 1
imr->umem = umem
num_pending_prefetch_inc()
if (live)
atomic_inc(num_pending_prefetch)
atomic_set(num_pending_prefetch,0) // Overwrites other thread's store
Further, live is being used with SRCU as the 'update' in an
acquire/release fashion, so it can not be read and written raw.
Move all live = 1's to after MR initialization is completed and use
smp_store_release/smp_load_acquire() for manipulating it.
Add a missing live = 0 when an implicit MR child is deleted, before
queuing work to do synchronize_srcu().
The barriers in update_odp_mr() were some broken attempt to create a
acquire/release, but were not even applied consistently and missed the
point, delete it as well.
Fixes: 6aec21f6a8 ("IB/mlx5: Page faults handling infrastructure")
Link: https://lore.kernel.org/r/20191001153821.23621-6-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
During destroy setting live = 0 and then synchronize_srcu() prevents
num_pending_prefetch from incrementing, and also, ensures that all work
holding that count is queued on the WQ. Testing before causes races of the
form:
CPU0 CPU1
dereg_mr()
mlx5_ib_advise_mr_prefetch()
srcu_read_lock()
num_pending_prefetch_inc()
if (!live)
live = 0
atomic_read() == 0
// skip flush_workqueue()
atomic_inc()
queue_work();
srcu_read_unlock()
WARN_ON(atomic_read()) // Fails
Swap the order so that the synchronize_srcu() prevents this.
Fixes: a6bc3875f1 ("IB/mlx5: Protect against prefetch of invalid MR")
Link: https://lore.kernel.org/r/20191001153821.23621-5-jgg@ziepe.ca
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
