The driver exposes interfaces that directly relate to HW state.
Upon fatal error, consumers of these interfaces (ULPs) that rely
on completion of all their posted work-request could hang, thereby
introducing dependencies in shutdown order. To prevent this from
happening, we manage the relevant resources (CQs, QPs) that are used
by the device. Upon a fatal error, we now generate simulated
completions for outstanding WQEs that were not completed at the
time the HW was reset.
It includes invoking the completion event handler for all involved
CQs so that the ULPs will poll those CQs. When polled we return
simulated CQEs with IB_WC_WR_FLUSH_ERR return code enabling ULPs
to clean up their resources and not wait forever for completions
upon receiving remove_one.
The above change requires an extra check in the data path to make
sure that when device is in error state, the simulated CQEs will
be returned and no further WQEs will be posted.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The SRP initiator allows to set max_sectors to a value that exceeds
the largest amount of data that can be mapped at once with an mlx4
HCA using fast registration and a page size of 4 KB. Hence modify
ib_map_mr_sg() such that it can map partial sg-elements. If an
sg-element has been mapped partially, let the caller know
which fraction has been mapped by adjusting *sg_offset.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Tested-by: Laurence Oberman <loberman@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Allocate proper context for arbitrary scatterlist registration
If ib_alloc_mr is called with IB_MR_MAP_ARB_SG, the driver
allocate a private klm list instead of a private page list.
Set the UMR wqe correctly when posting the fast registration.
Also, expose device cap IB_DEVICE_MAP_ARB_SG according to the
device id (until we have a FW bit that correctly exposes it).
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
These three related functions can't agree whether to put the
umrwr on the stack dirty and then memset it, or to initialize
it on the stack. Make them all agree.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Simplifies the code, and makes it more fair vs other users by using a
softirq for polling.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Haggai Eran <haggaie@mellanox.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch adds user-space support for memory windows allocation and
deallocation. It also exposes the supported types via
query_device_caps verb.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Tested-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Mlx5's mkey mechanism is also used for memory windows.
The current code base uses MR (memory region) naming, which is
inaccurate. Changing MR to mkey in order to represent its different
usages more accurately.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch adds support for re-registration of memory regions in MLX5.
The functionality is basically the same as deregister followed by
register, but attempts to reuse the existing resources as much as
possible.
Original memory keys are kept if possible, saving the need to
communicate new ones to remote peers.
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In order to add re-registration of memory region, some logic was
extracted to separate functions:
- ODP related logic.
- Some of the UMR WQE preparation code.
- DMA mapping.
- Umem creation.
- Creating MKey using FW interface.
- MR fields assignments after successful creation.
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The remove_keys() logic is performed as garbage collection task. Such
task is intended to be run when no other active processes are running.
The need_resched() will return TRUE if there are user tasks to be
activated in near future.
In such case, we don't execute remove_keys() and postpone
the garbage collection work to try to run in next cycle,
in order to free CPU resources to other tasks.
The possible pseudo-code to trigger such scenario:
1. Allocate a lot of MR to fill the cache above the limit.
2. Wait a small amount of time "to calm" the system.
3. Start CPU extensive operations on multi-node cluster.
4. Expect performance degradation during MR cache shrink operation.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
No ULP uses it anymore, go ahead and remove it.
Keep only the local invalidate part of the handlers.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Support the new memory registration API by allocating a
private page list array in mlx5_ib_mr and populate it when
mlx5_ib_map_mr_sg is invoked. Also, support IB_WR_REG_MR
by setting the exact WQE as IB_WR_FAST_REG_MR, just take the
needed information from different places:
- page_size, iova, length, access flags (ib_mr)
- page array (mlx5_ib_mr)
- key (ib_reg_wr)
The IB_WR_FAST_REG_MR handlers will be removed later when
all the ULPs will be converted.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch split up struct ib_send_wr so that all non-trivial verbs
use their own structure which embedds struct ib_send_wr. This dramaticly
shrinks the size of a WR for most common operations:
sizeof(struct ib_send_wr) (old): 96
sizeof(struct ib_send_wr): 48
sizeof(struct ib_rdma_wr): 64
sizeof(struct ib_atomic_wr): 96
sizeof(struct ib_ud_wr): 88
sizeof(struct ib_fast_reg_wr): 88
sizeof(struct ib_bind_mw_wr): 96
sizeof(struct ib_sig_handover_wr): 80
And with Sagi's pending MR rework the fast registration WR will also be
down to a reasonable size:
sizeof(struct ib_fastreg_wr): 64
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> [srp, srpt]
Reviewed-by: Chuck Lever <chuck.lever@oracle.com> [sunrpc]
Tested-by: Haggai Eran <haggaie@mellanox.com>
Tested-by: Sagi Grimberg <sagig@mellanox.com>
Tested-by: Steve Wise <swise@opengridcomputing.com>
The mlx5_ib_reg_user_mr() function will attempt to call clean_mr() in
its error flow even though there is never a case where the error flow
occurs with a valid MR pointer to destroy.
Remove the clean_mr() call and the incorrect comment above it.
Fixes: b4cfe447d4 ("IB/mlx5: Implement on demand paging by adding
support for MMU notifiers")
Cc: Eli Cohen <eli@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The pd now has a local_dma_lkey member which completely replaces
ib_get_dma_mr, use it instead.
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Use ib_alloc_mr with specific parameters.
Change the existing callers.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This was added in a thought of uniting all mr allocation
and deallocation routines but the fact is we have a single
deallocation routine already, ib_dereg_mr.
And, move mlx5_ib_destroy_mr specific logic into mlx5_ib_dereg_mr
(includes only signature stuff for now).
And, fixup the only callers (iser/isert) accordingly.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The only place that assigns mr inside the loop already does a break.
So "if (mr)" will never be true here since the function initializes mr
to NULL at the top. We can just drop the extra if and break here.
Signed-off-by: Roland Dreier <roland@purestorage.com>
Acked-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
- Query all supported types of dev caps on driver load.
- Store the Cap data outbox per cap type into driver private data.
- Introduce new Macros to access/dump stored caps (using the auto
generated data types).
- Obsolete SW representation of dev caps (no need for SW copy for each
cap).
- Modify IB driver to use new macros for checking caps.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we create an MR using reg_create, the mlx5_ib_dev pointer is not
updated on the new MR. This results in a kernel panics for ODP MRs
while handling page faults, when the mlx5_ib_update_mtt function uses
the invalid device pointer.
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
* Implement the relevant invalidation functions (zap MTTs as needed)
* Implement interlocking (and rollback in the page fault handlers) for
cases of a racing notifier and fault.
* With this patch we can now enable the capability bits for supporting RC
send/receive/RDMA read/RDMA write, and UD send.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
* Refactor MR registration and cleanup, and fix reg_pages accounting.
* Create a work queue to handle page fault events in a kthread context.
* Register a fault handler to get events from the core for each QP.
The registered fault handler is empty in this patch, and only a later
patch implements it.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The new function allows updating the page tables of a memory region
after it was created. This can be used to handle page faults and page
invalidations.
Since mlx5_ib_update_mtt will need to work from within page invalidation,
so it must not block on memory allocation. It employs an atomic memory
allocation mechanism that is used as a fallback when kmalloc(GFP_ATOMIC) fails.
In order to reuse code from mlx5_ib_populate_pas, the patch splits
this function and add the needed parameters.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This patch wraps together several changes needed for on-demand paging support
in the mlx5_ib_populate_pas function, and when registering memory regions.
* Instead of accepting a UMR bit telling the function to enable all
access flags, the function now accepts the access flags themselves.
* For on-demand paging memory regions, fill the memory tables from the
correct list, and enable/disable the access flags per-page according
to whether the page is present.
* A new bit is set to enable writing of access flags when using the
firmware create_mkey command.
* Disable contig pages when on-demand paging is enabled.
In addition the patch changes the UMR code to use PTR_ALIGN instead of
our own macro.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The current UMR interface doesn't allow partial updates to a memory
region's page tables. This patch changes the interface to allow that.
It also changes the way the UMR operation validates the memory
region's state. When set, IB_SEND_UMR_FAIL_IF_FREE will cause the UMR
operation to fail if the MKEY is in the free state. When it is
unchecked the operation will check that it isn't in the free state.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Since UMR code now uses its own context struct on the stack, the pas
and dma pointers for the UMR operation that remained in the mlx5_ib_mr
struct are not necessary. This patch removes them.
Fixes: a74d24168d ("IB/mlx5: Refactor UMR to have its own context struct")
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
If mlx5_core_create_mkey fails, decrease the pending counter to undo the
previous increment.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for a new mlx5 device which is VPI (i.e., ports can be
either IB or ETH), move the pci device functionality from mlx5_ib
to mlx5_core.
This involves the following changes:
1. Move mlx5_core_dev struct out of mlx5_ib_dev. mlx5_core_dev
is now an independent structure maintained by mlx5_core.
mlx5_ib_dev now has a pointer to that struct.
This requires changing a lot of places where the core_dev
struct was accessed via mlx5_ib_dev (now, this needs to
be a pointer dereference).
2. All PCI initializations are now done in mlx5_core. Thus,
it is now mlx5_core which does pci_register_device (and not
mlx5_ib, as was previously).
3. mlx5_ib now registers itself with mlx5_core as an "interface"
driver. This is very similar to the mechanism employed for
the mlx4 (ConnectX) driver. Once the HCA is initialized
(by mlx5_core), it invokes the interface drivers to do
their initializations.
4. There is a new event handler which the core registers:
mlx5_core_event(). This event handler invokes the
event handlers registered by the interfaces.
Based on a patch by Eli Cohen <eli@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to cast wr_id to unsigned long before casting to a pointer.
This fixes:
drivers/infiniband/hw/mlx5/mr.c: In function 'mlx5_umr_cq_handler':
>> drivers/infiniband/hw/mlx5/mr.c:724:13: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
context = (struct mlx5_ib_umr_context *)wc.wr_id;
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Instead of having the UMR context part of each memory region, allocate
a struct on the stack. This allows queuing multiple UMRs that access
the same memory region.
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The patch stores iova, pd and size during mr creation and after UMRs
that modify them. It removes the unused access flags field.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
For memory regions that are allocated using reg_umr, the suffix of
mlx5_core_create_mkey isn't being called. Instead the creation is
completed in a callback function (reg_mr_callback). This means that
these MRs aren't being added to the MR radix tree. Add them in the
callback.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
If ib_post_send fails when posting the UMR work request in reg_umr,
the code doesn't release the temporary pas buffer allocated, and
doesn't dma_unmap it.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This commit takes care of the generated signature error CQE generated
by the HW (if happened). The underlying mlx5 driver will handle
signature error completions and will mark the relevant memory region
as dirty.
Once the consumer gets the completion for the transaction, it must
check for signature errors on signature memory region using a new
lightweight verb ib_check_mr_status().
In case the user doesn't check for signature error (i.e. doesn't call
ib_check_mr_status() with status check IB_MR_CHECK_SIG_STATUS), the
memory region cannot be used for another signature operation
(REG_SIG_MR work request will fail).
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Support create_mr and destroy_mr verbs. Creating ib_mr may be done
for either ib_mr that will register regular page lists like
alloc_fast_reg_mr routine, or indirect ib_mrs that can register other
(pre-registered) ib_mrs in an indirect manner.
In addition user may request signature enable, that will mean that the
created ib_mr may be attached with signature attributes (BSF, PSVs).
Currently we only allow direct/indirect registration modes.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The variable start in struct mlx5_ib_mr is never used. Remove it.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The value of the local variable index is never used in reg_mr_callback().
Signed-off-by: Eli Cohen <eli@mellanox.com>
[ Remove now-unused variable delta too. - Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
In mlx5_mr_cache_init() the size variable is not used so remove it to
avoid compiler warnings when running with make W=1.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use asynchronous commands to execute up to eight concurrent create MR
commands. This is to fill memory caches faster so we keep consuming
from there. Also, increase timeout for shrinking caches to five
minutes.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Call mlx5_ib_populate_pas() before mapping the DMA buffer to ensure
the hardware reads the values written by the CPU.
Found by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The hardware requires that gather buffers for UMR work requests be
aligned to 2K.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Change the logic so we do not allocate memory nor map the device
before actually posting to the REG_UMR QP. In addition, unmap and free
the memory after we get completion.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Destroying the workqueue without flushing it first can lead to a case
in which the kernel tries to push a delayed work to the workqueue
which does not exist anymore.
Signed-off-by: Moshe Lazer <moshel@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
For copy_to/from_user() failure, the correct error code is -EFAULT not
-EPERM.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The driver is comprised of two kernel modules: mlx5_ib and mlx5_core.
This partitioning resembles what we have for mlx4, except that mlx5_ib
is the pci device driver and not mlx5_core.
mlx5_core is essentially a library that provides general functionality
that is intended to be used by other Mellanox devices that will be
introduced in the future. mlx5_ib has a similar role as any hardware
device under drivers/infiniband/hw.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
[ Merge in coccinelle fixes from Fengguang Wu <fengguang.wu@intel.com>.
- Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
