Mixing these together is just a mess, make a dedicated version,
mlx5_ib_update_mr_pas(), which directly loads the whole MTT for a non-ODP
MR.
The split out version can trivially use a simple loop with
rdma_for_each_block() which allows using the core code to compute the MR
pages and avoids seeking in the SGL list after each chunk as the
__mlx5_ib_populate_pas() call required.
Significantly speeds loading large MTTs.
Link: https://lore.kernel.org/r/20201026132314.1336717-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The memory allocation is quite complicated, and makes this function hard
to understand. Refactor things so that a function call sets up the WR, SG,
DMA mapping and buffer, further splitting that into buffer and DMA/wr.
This also slightly changes the buffer allocation logic to try an order 0
page allocation (with OOM warnings on) before going to the emergency page.
Link: https://lore.kernel.org/r/20201026132314.1336717-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
For the user MR path, instead of calling this after getting the umem, call
it as part of creating the struct mlx5_ib_mr and distill its output to a
single page_shift stored inside the mr.
This avoids passing around the tuple of its output. Based on the umem and
page_shift, the output arguments can be computed using:
count == ib_umem_num_pages(mr->umem)
shift == mr->page_shift
ncont == ib_umem_num_dma_blocks(mr->umem, 1 << mr->page_shift)
order == order_base_2(ncont)
And since mr->page_shift == umem_odp->page_shift then ncont ==
ib_umem_num_dma_blocks() == ib_umem_odp_num_pages() for ODP umems.
Link: https://lore.kernel.org/r/20201026131936.1335664-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
reg_pages should always contain mr->npage since when the mr is finally
de-reg'd it is always subtracted out.
If there were any error exits then mlx5_ib_rereg_user_mr() would leave the
reg_pages adjusted and this will cause it to be double subtracted
eventually.
The manipulation of reg_pages is inherently connected to the umem, so lift
it out of set_mr_fields() and only adjust it around creating/destroying a
umem.
reg_pages is only used for diagnostics in sysfs.
Fixes: 7d0cc6edcc ("IB/mlx5: Add MR cache for large UMR regions")
Link: https://lore.kernel.org/r/20201026131936.1335664-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The is only ever set to non-zero if the MR is from the cache, and if it is
cached then the order is in cached_ent->order.
Make it clearer that use_umr_mtt_update() only returns true for cached MRs
and remove the redundant data.
Link: https://lore.kernel.org/r/20201026131936.1335664-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Each driver should check that the QP attrs create_flags is supported.
Unfortuantely when create_flags was added to the QP attrs the drivers were
not updated. uverbs_ex_cmd_mask was used to block it - even though kernel
drivers use these flags too.
Check that flags is zero in all drivers that don't use it, remove
IB_USER_VERBS_EX_CMD_CREATE_QP from uverbs_ex_cmd_mask. Fix the error code
to be EOPNOTSUPP.
Link: https://lore.kernel.org/r/8-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Each driver should check that the CQ attrs is supported. Unfortuantely
when flags was added to the CQ attrs the drivers were not updated,
uverbs_ex_cmd_mask was used to block it. This was missed when create CQ
was converted to ioctl, so non-zero flags could have been passed into
drivers.
Check that flags is zero in all drivers that don't use it, remove
IB_USER_VERBS_EX_CMD_CREATE_CQ from uverbs_ex_cmd_mask.
Fixes: 41b2a71fc8 ("IB/uverbs: Move ioctl path of create_cq and destroy_cq to a new file")
Link: https://lore.kernel.org/r/7-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
These functions all depend on the driver providing a specific op:
- REREG_MR is rereg_user_mr(). bnxt_re set this without providing the op
- ATTACH/DEATCH_MCAST is attach_mcast()/detach_mcast(). usnic set this
without providing the op
- OPEN_QP doesn't involve the driver but requires a XRCD. qedr provides
xrcd but forgot to set it, usnic doesn't provide XRCD but set it anyhow.
- OPEN/CLOSE_XRCD are the ops alloc_xrcd()/dealloc_xrcd()
- CREATE_SRQ/DESTROY_SRQ are the ops create_srq()/destroy_srq()
- QUERY/MODIFY_SRQ is op query_srq()/modify_srq(). hns sets this but
sometimes supplies a NULL op.
- RESIZE_CQ is op resize_cq(). bnxt_re sets this boes doesn't supply an op
- ALLOC/DEALLOC_MW is alloc_mw()/dealloc_mw(). cxgb4 provided an
(now deleted) implementation but no userspace
All drivers were checked that no drivers provide the op without also
setting uverbs_cmd_mask so this should have no functional change.
Link: https://lore.kernel.org/r/4-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Since a while now the uverbs layer checks if the driver implements a
function before allowing the ucmd to proceed. This largely obsoletes the
cmd_mask stuff, but there is some tricky bits in drivers preventing it
from being removed.
Remove the easy elements of uverbs_ex_cmd_mask by pre-setting them in the
core code. These are triggered soley based on the related ops function
pointer.
query_device_ex is not triggered based on an op, but all drivers already
implement something compatible with the extension, so enable it globally
too.
Link: https://lore.kernel.org/r/2-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
When a mlx5 core devlink instance is reloaded in different net namespace,
its associated IB device is deleted and recreated.
Example sequence is:
$ ip netns add foo
$ devlink dev reload pci/0000:00:08.0 netns foo
$ ip netns del foo
mlx5 IB device needs to attach and detach the netdevice to it through the
netdev notifier chain during load and unload sequence. A below call graph
of the unload flow.
cleanup_net()
down_read(&pernet_ops_rwsem); <- first sem acquired
ops_pre_exit_list()
pre_exit()
devlink_pernet_pre_exit()
devlink_reload()
mlx5_devlink_reload_down()
mlx5_unload_one()
[...]
mlx5_ib_remove()
mlx5_ib_unbind_slave_port()
mlx5_remove_netdev_notifier()
unregister_netdevice_notifier()
down_write(&pernet_ops_rwsem);<- recurrsive lock
Hence, when net namespace is deleted, mlx5 reload results in deadlock.
When deadlock occurs, devlink mutex is also held. This not only deadlocks
the mlx5 device under reload, but all the processes which attempt to
access unrelated devlink devices are deadlocked.
Hence, fix this by mlx5 ib driver to register for per net netdev notifier
instead of global one, which operats on the net namespace without holding
the pernet_ops_rwsem.
Fixes: 4383cfcc65 ("net/mlx5: Add devlink reload")
Link: https://lore.kernel.org/r/20201026134359.23150-1-parav@nvidia.com
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Separate IB_GID_TYPE_IB and IB_GID_TYPE_ROCE to two different values, so
enum ib_gid_type will match the gid types of the new query GID table API
which will be introduced in the following patches.
This change in enum ib_gid_type requires to change also enum
rdma_network_type by separating RDMA_NETWORK_IB and RDMA_NETWORK_ROCE_V1
values.
Link: https://lore.kernel.org/r/20200923165015.2491894-3-leon@kernel.org
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Enable ODP sync without faulting, this improves performance by reducing
the number of page faults in the system.
The gain from this option is that the device page table can be aligned
with the presented pages in the CPU page table without causing page
faults.
As of that, the overhead on data path from hardware point of view to
trigger a fault which end-up by calling the driver to bring the pages
will be dropped.
Link: https://lore.kernel.org/r/20200930163828.1336747-3-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Move to use hmm_range_fault() instead of get_user_pags_remote() to improve
performance in a few aspects:
This includes:
- Dropping the need to allocate and free memory to hold its output
- No need any more to use put_page() to unpin the pages
- The logic to detect contiguous pages is done based on the returned
order, no need to run per page and evaluate.
In addition, moving to use hmm_range_fault() enables to reduce page faults
in the system with it's snapshot mode, this will be introduced in next
patches from this series.
As part of this, cleanup some flows and use the required data structures
to work with hmm_range_fault().
Link: https://lore.kernel.org/r/20200930163828.1336747-2-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The restrack is going to manage memory of all IB objects and must be
called before object is created. GSI QP in the mlx5_ib separated between
creating dummy interface and HW object beneath. This was achieved by
double call to ib_create_qp().
In order to skip such reentry call to internal driver create_qp code.
Link: https://lore.kernel.org/r/20200922091106.2152715-3-leon@kernel.org
Reviewed-by: Mark Zhang <markz@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Once a mkey is created it can be modified using UMR. This is desirable for
performance reasons. However, different hardware has restrictions on what
modifications are possible using UMR. Make sense of these checks:
- mlx5_ib_can_reconfig_with_umr() returns true if the access flags can be
altered. Most cases create MRs using 0 access flags (now made clear by
consistent use of set_mkc_access_pd_addr_fields()), but the old logic
here was tormented. Make it clear that this is checking if the current
access_flags can be modified using UMR to different access_flags. It is
always OK to use UMR to change flags that all HW supports.
- mlx5_ib_can_load_pas_with_umr() returns true if UMR can be used to
enable and update the PAS/XLT. Enabling requires updating the entity
size, so UMR ends up completely disabled on this old hardware. Make it
clear why it is disabled. FRWR, ODP and cache always requires
mlx5_ib_can_load_pas_with_umr().
- mlx5_ib_pas_fits_in_mr() is used to tell if an existing MR can be
resized to hold a new PAS list. This only works for cached MR's because
we don't store the PAS list size in other cases.
To be very clear, arrange things so any pre-created MR's in the cache
check the newly requested access_flags before allowing the MR to leave the
cache. If UMR cannot set the required access_flags the cache fails to
create the MR.
This in turn means relaxed ordering and atomic are now correctly blocked
early for implicit ODP on older HW.
Link: https://lore.kernel.org/r/20200914112653.345244-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
set_reg_wr() always fails if !umr_modify_entity_size_disabled because
mlx5_ib_can_use_umr() always fails. Without set_reg_wr() IB_WR_REG_MR
doesn't work and that means the device should not advertise
IB_DEVICE_MEM_MGT_EXTENSIONS.
Fixes: 841b07f99a ("IB/mlx5: Block MR WR if UMR is not possible")
Link: https://lore.kernel.org/r/20200914112653.345244-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Any mkey that is not enabled and assigned to userspace should have the PD
set to a kernel owned PD.
When cache entries are created for the first time the PDN is set to 0,
which is probably a kernel PD, but be explicit.
When a MR is registered using the hybrid reg_create with UMR xlt & enable
the disabled mkey is pointing at the user PD, keep it pointing at the
kernel until a UMR enables it and sets the user PD.
Fixes: 9ec4483a3f ("IB/mlx5: Move MRs to a kernel PD when freeing them to the MR cache")
Link: https://lore.kernel.org/r/20200914112653.345244-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Leon Romanovsky says:
====================
This series from Alex extends software steering interface to support
devices with extra capability "sw_owner_2" which will replace existing
"sw_owner".
====================
Based on the mlx5-next branch at
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux
due to dependencies.
* branch 'mlx5_sw_owner_v2:
RDMA/mlx5: Expose TIR and QP ICM address for sw_owner_v2 devices
RDMA/mlx5: Allow DM allocation for sw_owner_v2 enabled devices
RDMA/mlx5: Add sw_owner_v2 bit capability
Leon Romanovsky says:
====================
IBTA declares speed as 16 bits, but kernel stores it in u8. This series
fixes in-kernel declaration while keeping external interface intact.
====================
Based on the mlx5-next branch at
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux
due to dependencies.
* branch 'mlx5_active_speed':
RDMA: Fix link active_speed size
RDMA/mlx5: Delete duplicated mlx5_ptys_width enum
net/mlx5: Refactor query port speed functions
The functions mlx5_query_port_link_width_oper and
mlx5_query_port_ib_proto_oper are always called together, so combine them
to a new function called mlx5_query_port_oper to avoid duplication.
And while the mlx5i_get_port_settings is the same as
mlx5_query_port_oper therefore let's remove it.
According to the IB spec link_width_oper and ib_proto_oper should be u16
and not as written u8, so perform casting as a preparation to cross-RDMA
patch which will fix that type for all drivers in the RDMA subsystem.
Fixes: ada68c31ba ("net/mlx5: Introduce a new header file for physical port functions")
Signed-off-by: Aharon Landau <aharonl@mellanox.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
ib_umem_num_pages() should only be used by things working with the SGL in
CPU pages directly.
Drivers building DMA lists should use the new ib_num_dma_blocks() which
returns the number of blocks rdma_umem_for_each_block() will return.
To make this general for DMA drivers requires a different implementation.
Computing DMA block count based on umem->address only works if the
requested page size is < PAGE_SIZE and/or the IOVA == umem->address.
Instead the number of DMA pages should be computed in the IOVA address
space, not umem->address. Thus the IOVA has to be stored inside the umem
so it can be used for these calculations.
For now set it to umem->address by default and fix it up if
ib_umem_find_best_pgsz() was called. This allows drivers to be converted
to ib_umem_num_dma_blocks() safely.
Link: https://lore.kernel.org/r/6-v2-270386b7e60b+28f4-umem_1_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
