Currently the rxe driver uses red-black trees to add indices to the rxe
object pools. Linux xarrays provide a better way to implement the same
functionality for indices. This patch replaces red-black trees by xarrays
for pool objects. Since xarrays already have a spinlock use that in place
of the pool rwlock. Make sure that all changes in the xarray(index) and
kref(ref counnt) occur atomically.
Link: https://lore.kernel.org/r/20220304000808.225811-9-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
There is only one remaining object type that allocates its own memory,
that is mr. So the sense of RXE_POOL_NO_ALLOC is changed to
RXE_POOL_ALLOC. Add checks to rxe_alloc() and rxe_add_to_pool() to make
sure the correct call is used for the setting of this flag.
Link: https://lore.kernel.org/r/20220304000808.225811-4-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
A previous patch replaced all irqsave locks in rxe with bh locks. This
ran into problems because rdmacm has a bad habit of calling rdma verbs
APIs while disabling irqs. This is not allowed during spin_unlock_bh()
causing programs that use rdmacm to fail. This patch reverts the changes
to locks that had this problem or got dragged into the same mess. After
this patch blktests/check -q srp now runs correctly.
Link: https://lore.kernel.org/r/20220215194448.44369-1-rpearsonhpe@gmail.com
Fixes: 21adfa7a3c ("RDMA/rxe: Replace irqsave locks with bh locks")
Reported-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Reported-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Tested-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In rxe_pool.c currently there are many cases where it is necessary to
compute the offset from a pool element struct to the object containing it
in a type independent way where the offset is different for each type. By
saving a pointer to the object when they are created extra work can be
saved.
Link: https://lore.kernel.org/r/20211103050241.61293-5-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Most of the locks in the rxe driver are _irqsave/restore locks but in fact
there are no interrupt threads that run rxe code or share data with
rxe. There are softirq threads and data sharing so the appropriate lock
type is _bh. This patch replaces all irqsave type locks with bh type
locks.
Link: https://lore.kernel.org/r/20211103050241.61293-2-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Make struct rxe_type_info static const and local to the only uses. Moves
a bit of data to text.
$ size drivers/infiniband/sw/rxe/rxe_pool.o* (defconfig w/ infiniband swe)
text data bss dec hex filename
4456 12 0 4468 1174 drivers/infiniband/sw/rxe/rxe_pool.o.new
3817 652 0 4469 1175 drivers/infiniband/sw/rxe/rxe_pool.o.old
Link: https://lore.kernel.org/r/166b715d71f98336e8ecab72b0dbdd266eee9193.camel@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In the original rxe implementation it was intended to use a common object
to represent MRs and MWs but they are different enough to separate these
into two objects.
This allows replacing the mem name with mr for MRs which is more
consistent with the style for the other objects and less likely to be
confusing. This is a long patch that mostly changes mem to mr where it
makes sense and adds a new rxe_mw struct.
Link: https://lore.kernel.org/r/20210325212425.2792-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Acked-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
rxe_pool.c uses the field pool->state to mark a pool as invalid when it is
shut down and checks it in several pool APIs to verify that the pool has
not been shut down. This is unneeded because the pools are not marked
invalid unless the entire driver is being removed at which point no
functional APIs should or could be executing. This patch removes this
field and associated code.
Link: https://lore.kernel.org/r/20210125211641.2694-6-rpearson@hpe.com
Suggested-by: zyjzyj2000@gmail.c
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
rxe_pool.c takes references to the pool and ib_device structs for each
object allocated and also keeps an atomic num_elem count in each
pool. This is more work than is needed. Pool allocation is only called
from verbs APIs which already have references to ib_device and pools are
only diasbled when the driver is removed so no protection of the pool
addresses are needed. The elem count is used to warn if elements are still
present in a pool when it is cleaned up which is useful.
This patch eliminates the references to the ib_device and pool structs.
Link: https://lore.kernel.org/r/20210125211641.2694-5-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
rxe_alloc() used the RXE_POOL_ATOMIC flag in rxe_type_info to select
GFP_ATOMIC in calls to kzalloc(). This was intended to handle cases where
an object could be created in interrupt context. This no longer occurs
since allocating those objects has moved into the core so this flag is not
necessary. An incorrect use of this flag was still present for rxe_mc_elem
objects and is removed.
Link: https://lore.kernel.org/r/20210125211641.2694-4-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
A recent patch which added an 'unlocked' version of rxe_alloc introduced a
bug causing kzalloc(..., GFP_KERNEL) to be called while holding a spin
lock. This patch corrects that error.
rxe_alloc_nl() should always be called while holding the pool->pool_lock
so the 2nd argument to kzalloc there should be GFP_ATOMIC.
rxe_alloc() prior to the change only locked the code around checking that
pool->state is RXE_POOL_STATE_VALID to avoid races between working threads
and a thread shutting down the rxe driver. This patch reverts rxe_alloc()
to this behavior so the lock is not held when kzalloc() is called.
Link: https://lore.kernel.org/r/20210125211641.2694-2-rpearson@hpe.com
Reported-by: syzbot+ec2fd72374785d0e558e@syzkaller.appspotmail.com
Fixes: 3853c35e24 ("RDMA/rxe: Add unlocked versions of pool APIs")
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The existing pool APIs use the rw_lock pool_lock to protect critical
sections that change the pool state. This does not correctly implement a
typical sequence like the following
elem = <lookup key in pool>
if found use elem else
elem = <alloc new elem in pool>
<add key to elem>
Which is racy if multiple threads are attempting to perform this at the
same time. We want the second thread to use the elem created by the first
thread not create two equivalent elems.
This patch adds new APIs that are the same as existing APIs but do not
take the pool_lock. A caller can then take the lock and perform a sequence
of pool operations and then release the lock.
Link: https://lore.kernel.org/r/20201216231550.27224-7-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Replace 'void *' parameters with 'struct rxe_pool_entry *' and use a macro
to allow:
rxe_add_index,
rxe_drop_index,
rxe_add_key,
rxe_drop_key and
rxe_add_to_pool
APIs to be type safe against changing the position of pelem in the
objects.
Link: https://lore.kernel.org/r/20201216231550.27224-6-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The allocate, lookup index, lookup key and cleanup routines in rxe_pool.c
currently are not type safe against relocating the pelem field in the
objects. Planned changes to move allocation of objects into rdma-core make
addressing this a requirement.
Use the elem_offset field in rxe_type_info make these APIs safe against
moving the pelem field.
Link: https://lore.kernel.org/r/20201216231550.27224-5-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The rxe verbs objects each include an rdma-core object 'ib_xxx'
and a rxe_pool_entry 'pelem' in addition to rxe specific data.
Originally these all had pelem first and ib_xxx second. Currently
about half have ib_xxx first and half have pelem first. Saving
the offset of the pelem field in rxe_type info will enable making
the rxe_pool APIs type safe as the pelem field continues to vary.
Link: https://lore.kernel.org/r/20201216231550.27224-4-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Change rxe pools to use kzalloc instead of kmem_cache to allocate memory
for rxe objects. The pools are not really necessary and they trigger
hardened user copy warnings as the ioctl framework copies the QP number
directly to userspace.
Also the general project to move object alloation to the core code will
eventually clean these out anyhow.
Link: https://lore.kernel.org/r/20200827163535.2632-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Convert SRQ allocation from drivers to be in the IB/core
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Simplify drivers by ensuring lifetime of ib_ah object. The changes
in .create_ah() go hand in hand with relevant update in .destroy_ah().
We will use this opportunity and convert .destroy_ah() to don't fail, as
it was suggested a long time ago, because there is nothing to do in case
of failure during destroy.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Following the PD conversion patch, do the same for ucontext allocations.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
rxe does not have correct locking for its registration/unregistration
paths, use the core code to handle it instead. In this mode
ib_unregister_device will also do the dealloc, so rxe is required to do
clean up from a callback.
The core code ensures that unregistration is done only once, and generally
takes care of locking and concurrency problems for rxe.
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The PD allocations in IB/core allows us to simplify drivers and their
error flows in their .alloc_pd() paths. The changes in .alloc_pd() go hand
in had with relevant update in .dealloc_pd().
We will use this opportunity and convert .dealloc_pd() to don't fail, as
it was suggested a long time ago, failures are not happening as we have
never seen a WARN_ON print.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
