Introduce a generic APIs to iterate over all the devices which are part
of the LAG. This API replace mlx5_lag_get_peer_mdev() which retrieve
only a single peer device from the lag.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The cited patch introduce ib port for the slave device uplink in
case of multiport eswitch. However, this ib port didn't perform
anything when unloaded.
Unload the new ib port properly.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
After commit 6d758147c7 ("RDMA/bnxt_re: Use auxiliary driver interface")
the active_{speed, width} attributes are reported incorrectly, This is
happening because ib_get_eth_speed() is called only once from
bnxt_re_ib_init() - Fix this issue by calling ib_get_eth_speed() from
bnxt_re_query_port().
Fixes: 6d758147c7 ("RDMA/bnxt_re: Use auxiliary driver interface")
Link: https://lore.kernel.org/r/20230529153525.87254-1-kheib@redhat.com
Signed-off-by: Kamal Heib <kheib@redhat.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Add kref_read() of mmu_rb_node.refcount in hfi1_mmu_rb_template-type
tracepoint output.
Change hfi1_mmu_rb_template tracepoint to take a struct mmu_rb_node* and
record the values it needs from that. This makes the trace_hfi1_mmu*()
calls shorter and easier to read.
Add hfi1_mmu_release_node() tracepoint before all
mmu_rb_node->handler->ops->remove() calls.
Make hfi1_mmu_rb_search() tracepoint its own tracepoint type separate
from hfi1_mmu_rb_template since hfi1_mmu_rb_search() does not take a
struct mmu_rb_node*.
Link: https://lore.kernel.org/r/168451525987.3702129.12824880387615916700.stgit@awfm-02.cornelisnetworks.com
Reviewed-by: Dean Luick <dean.luick@cornelisnetworks.com>
Signed-off-by: Brendan Cunningham <bcunningham@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The hfi1 user SDMA pinned-page cache will leave a stale cache entry when
the cache-entry's virtual address range is invalidated but that cache
entry is in-use by an outstanding SDMA request.
Subsequent user SDMA requests with buffers in or spanning the virtual
address range of the stale cache entry will result in packets constructed
from the wrong memory, the physical pages pointed to by the stale cache
entry.
To fix this, remove mmu_rb_node cache entries from the mmu_rb_handler
cache independent of the cache entry's refcount. Add 'struct kref
refcount' to struct mmu_rb_node and manage mmu_rb_node lifetime with
kref_get() and kref_put().
mmu_rb_node.refcount makes sdma_mmu_node.refcount redundant. Remove
'atomic_t refcount' from struct sdma_mmu_node and change sdma_mmu_node
code to use mmu_rb_node.refcount.
Move the mmu_rb_handler destructor call after a
wait-for-SDMA-request-completion call so mmu_rb_nodes that need
mmu_rb_handler's workqueue to queue themselves up for destruction from an
interrupt context may do so.
Fixes: f48ad614c1 ("IB/hfi1: Move driver out of staging")
Fixes: 00cbce5cbf ("IB/hfi1: Fix bugs with non-PAGE_SIZE-end multi-iovec user SDMA requests")
Link: https://lore.kernel.org/r/168451393605.3700681.13493776139032178861.stgit@awfm-02.cornelisnetworks.com
Reviewed-by: Dean Luick <dean.luick@cornelisnetworks.com>
Signed-off-by: Brendan Cunningham <bcunningham@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
There is a reference count error in error path code and a potential race
in check_rkey() in rxe_resp.c. When looking up the rkey for a memory
window the reference to the mw from rxe_lookup_mw() is dropped before a
reference is taken on the mr referenced by the mw. If the mr is destroyed
immediately after the call to rxe_put(mw) the mr pointer is unprotected
and may end up pointing at freed memory. The rxe_get(mr) call should take
place before the rxe_put(mw) call.
All errors in check_rkey() call rxe_put(mw) if mw is not NULL but it was
already called after the above. The mw pointer should be set to NULL after
the rxe_put(mw) call to prevent this from happening.
Fixes: cdd0b85675 ("RDMA/rxe: Implement memory access through MWs")
Link: https://lore.kernel.org/r/20230517211509.1819998-1-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In rxe_net.c a received packet, from udp or loopback, is passed to
rxe_rcv() in rxe_recv.c as a udp packet. I.e. skb->data is pointing at the
udp header. But rxe_rcv() makes length checks to verify the packet is long
enough to hold the roce headers as if it were a roce
packet. I.e. skb->data pointing at the bth header. A runt packet would
appear to have 8 more bytes than it actually does which may lead to
incorrect behavior.
This patch calls skb_pull() to adjust the skb to point at the bth header
before calling rxe_rcv() which fixes this error.
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Link: https://lore.kernel.org/r/20230517172242.1806340-1-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In current design:
1. PD and clt_path->s.dev are shared among connections.
2. every con[n]'s cleanup phase will call destroy_con_cq_qp()
3. clt_path->s.dev will be always decreased in destroy_con_cq_qp(), and
when clt_path->s.dev become zero, it will destroy PD.
4. when con[1] failed to create, con[1] will not take clt_path->s.dev,
but it try to decreased clt_path->s.dev
So, in case create_cm(con[0]) succeeds but create_cm(con[1]) fails,
destroy_con_cq_qp(con[1]) will be called first which will destroy the PD
while this PD is still taken by con[0].
Here, we refactor the error path of create_cm() and init_conns(), so that
we do the cleanup in the order they are created.
The warning occurs when destroying RXE PD whose reference count is not
zero.
rnbd_client L597: Mapping device /dev/nvme0n1 on session client, (access_mode: rw, nr_poll_queues: 0)
------------[ cut here ]------------
WARNING: CPU: 0 PID: 26407 at drivers/infiniband/sw/rxe/rxe_pool.c:256 __rxe_cleanup+0x13a/0x170 [rdma_rxe]
Modules linked in: rpcrdma rdma_ucm ib_iser rnbd_client libiscsi rtrs_client scsi_transport_iscsi rtrs_core rdma_cm iw_cm ib_cm crc32_generic rdma_rxe udp_tunnel ib_uverbs ib_core kmem device_dax nd_pmem dax_pmem nd_vme crc32c_intel fuse nvme_core nfit libnvdimm dm_multipath scsi_dh_rdac scsi_dh_emc scsi_dh_alua dm_mirror dm_region_hash dm_log dm_mod
CPU: 0 PID: 26407 Comm: rnbd-client.sh Kdump: loaded Not tainted 6.2.0-rc6-roce-flush+ #53
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:__rxe_cleanup+0x13a/0x170 [rdma_rxe]
Code: 45 84 e4 0f 84 5a ff ff ff 48 89 ef e8 5f 18 71 f9 84 c0 75 90 be c8 00 00 00 48 89 ef e8 be 89 1f fa 85 c0 0f 85 7b ff ff ff <0f> 0b 41 bc ea ff ff ff e9 71 ff ff ff e8 84 7f 1f fa e9 d0 fe ff
RSP: 0018:ffffb09880b6f5f0 EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff99401f15d6a8 RCX: 0000000000000000
RDX: 0000000000000001 RSI: ffffffffbac8234b RDI: 00000000ffffffff
RBP: ffff99401f15d6d0 R08: 0000000000000001 R09: 0000000000000001
R10: 0000000000002d82 R11: 0000000000000000 R12: 0000000000000001
R13: ffff994101eff208 R14: ffffb09880b6f6a0 R15: 00000000fffffe00
FS: 00007fe113904740(0000) GS:ffff99413bc00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007ff6cde656c8 CR3: 000000001f108004 CR4: 00000000001706f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<TASK>
rxe_dealloc_pd+0x16/0x20 [rdma_rxe]
ib_dealloc_pd_user+0x4b/0x80 [ib_core]
rtrs_ib_dev_put+0x79/0xd0 [rtrs_core]
destroy_con_cq_qp+0x8a/0xa0 [rtrs_client]
init_path+0x1e7/0x9a0 [rtrs_client]
? __pfx_autoremove_wake_function+0x10/0x10
? lock_is_held_type+0xd7/0x130
? rcu_read_lock_sched_held+0x43/0x80
? pcpu_alloc+0x3dd/0x7d0
? rtrs_clt_init_stats+0x18/0x40 [rtrs_client]
rtrs_clt_open+0x24f/0x5a0 [rtrs_client]
? __pfx_rnbd_clt_link_ev+0x10/0x10 [rnbd_client]
rnbd_clt_map_device+0x6a5/0xe10 [rnbd_client]
Fixes: 6a98d71dae ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/1682384563-2-4-git-send-email-lizhijian@fujitsu.com
Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Tested-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Commit df8fc4e934 ("kbuild: Enable -fstrict-flex-arrays=3") triggers a
warning for fortified memset():
In function 'fortify_memset_chk',
inlined from 'irdma_clr_wqes' at drivers/infiniband/hw/irdma/uk.c:103:4:
include/linux/fortify-string.h:493:25: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror=attribute-warning]
493 | __write_overflow_field(p_size_field, size);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The problem here isthat the inner array only has four 8-byte elements, so
clearing 4096 bytes overflows that. As this structure is part of an outer
array, change the code to pass a pointer to the irdma_qp_quanta instead,
and change the size argument for readability, matching the comment above
it.
Fixes: 551c46edc7 ("RDMA/irdma: Add user/kernel shared libraries")
Link: https://lore.kernel.org/r/20230523111859.2197825-1-arnd@kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
There is a window where the poll cq may use a QP that has been freed.
This can happen if a CQE is polled before irdma_clean_cqes() can clear the
CQE's related to the QP and the destroy QP races to free the QP memory.
then the QP structures are used in irdma_poll_cq. Fix this by moving the
clearing of CQE's before the reference is removed and the QP is destroyed.
Fixes: b48c24c2d7 ("RDMA/irdma: Implement device supported verb APIs")
Link: https://lore.kernel.org/r/20230522155654.1309-3-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
do_tcp_sendpages() is now just a small wrapper around tcp_sendmsg_locked(),
so inline it, allowing do_tcp_sendpages() to be removed. This is part of
replacing ->sendpage() with a call to sendmsg() with MSG_SPLICE_PAGES set.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Bernard Metzler <bmt@zurich.ibm.com>
Reviewed-by: Tom Talpey <tom@talpey.com>
cc: Jason Gunthorpe <jgg@ziepe.ca>
cc: Leon Romanovsky <leon@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
bnxt_re currently uses the names "bnxt_qplib_creq" and "bnxt_qplib_nq-0"
while registering IRQs. There is no way to distinguish the IRQs of
different device ports when there are multiple IB devices registered.
This could make the scenarios worse where one want to pin IRQs of a device
port to certain CPUs.
Fixed the code to use unique names which has PCI BDF information while
registering interrupts like: "bnxt_re-nq-0@pci:0000:65:00.0" and
"bnxt_re-creq@pci:0000:65:00.1".
Fixes: 1ac5a40479 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-4-git-send-email-selvin.xavier@broadcom.com
Reviewed-by: Bhargava Chenna Marreddy <bhargava.marreddy@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
When the ulp hook to start the IRQ fails because the rings are not
available, tasklets are not enabled. In this case when the driver is
unloaded, driver calls CREQ tasklet_kill. This causes an indefinite hang
as the tasklet is not enabled.
Driver shouldn't call tasklet_kill if it is not enabled. So using the
creq->requested and nq->requested flags to identify if both tasklets/irqs
are registered. Checking this flag while scheduling the tasklet from
ISR. Also, added a cleanup for disabling tasklet, in case request_irq
fails during start_irq.
Check for return value for bnxt_qplib_rcfw_start_irq and in case the
bnxt_qplib_rcfw_start_irq fails, return bnxt_re_start_irq without
attempting to start NQ IRQs.
Fixes: 1ac5a40479 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-2-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Long message loopback slice is used for achieving traffic balance between
QPs. It prevents the problem that QPs with large traffic occupying the
hardware pipeline for a long time and QPs with small traffic cannot be
scheduled.
Currently, its maximum value is set to 16K, which means only after a QP
sends 16K will the second QP be scheduled. This value is too large, which
will lead to unbalanced traffic scheduling, and thus it needs to be
modified.
The setting range of the long message loopback slice is modified to be
from 1024 (the lower limit supported by hardware) to mtu. Actual testing
shows that this value can significantly reduce error in hardware traffic
scheduling.
This solution is compatible with both HIP08 and HIP09. The modified
lp_pktn_ini has a maximum value of 2 (when mtu is 256), so the range
checking code for lp_pktn_ini is no longer necessary and needs to be
deleted.
Fixes: 0e60778efb ("RDMA/hns: Modify the value of MAX_LP_MSG_LEN to meet hardware compatibility")
Link: https://lore.kernel.org/r/20230512092245.344442-4-huangjunxian6@hisilicon.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
For hns, the specification of an entry like resource (E.g. WQE/CQE/EQE)
depends on BT page size, buf page size and hopnum. For user mode, the buf
page size depends on UMEM. Therefore, the actual specification is
controlled by BT page size and hopnum.
The current BT page size and hopnum are obtained from firmware. This makes
the driver inflexible and introduces unnecessary constraints. Resource
allocation failures occur in many scenarios.
This patch will calculate whether the BT page size set by firmware is
sufficient before allocating BT, and increase the BT page size if it is
insufficient.
Fixes: 1133401412 ("RDMA/hns: Optimize base address table config flow for qp buffer")
Link: https://lore.kernel.org/r/20230512092245.344442-3-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
On HIP08, the queried timeout attr is different from the timeout attr
configured by the user.
It is found by rdma-core testcase test_rdmacm_async_traffic:
======================================================================
FAIL: test_rdmacm_async_traffic (tests.test_rdmacm.CMTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
File "./tests/test_rdmacm.py", line 33, in test_rdmacm_async_traffic
self.two_nodes_rdmacm_traffic(CMAsyncConnection, self.rdmacm_traffic,
File "./tests/base.py", line 382, in two_nodes_rdmacm_traffic
raise(res)
AssertionError
Fixes: 926a01dc00 ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/20230512092245.344442-2-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Device uses 4KB size blocks for user pages indirect list while the
driver creates those blocks with the size of PAGE_SIZE of the kernel. On
kernels with PAGE_SIZE different than 4KB (ARM RHEL), this leads to a
failure on register MR with indirect list because of the miss
communication between driver and device.
Fixes: 40909f664d ("RDMA/efa: Add EFA verbs implementation")
Link: https://lore.kernel.org/r/20230511115103.13876-1-ynachum@amazon.com
Reviewed-by: Firas Jahjah <firasj@amazon.com>
Reviewed-by: Michael Margolin <mrgolin@amazon.com>
Signed-off-by: Yonatan Nachum <ynachum@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Driver populates the list of pages used for Memory region wrongly when
page size is more than system page size. This is causing a failure when
some of the applications that creates MR with page size as 2M. Since HW
can support multiple page sizes, pass the correct page size while creating
the MR.
Also, driver need not adjust the number of pages when HW Queues are
created with user memory. It should work with the number of dma blocks
returned by ib_umem_num_dma_blocks. Fix this calculation also.
Fixes: 0c4dcd6028 ("RDMA/bnxt_re: Refactor hardware queue memory allocation")
Fixes: f6919d5638 ("RDMA/bnxt_re: Code refactor while populating user MRs")
Link: https://lore.kernel.org/r/1683484169-9539-1-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Pull rdma updates from Jason Gunthorpe:
"Usual wide collection of unrelated items in drivers:
- Driver bug fixes and treewide cleanups in hfi1, siw, qib, mlx5,
rxe, usnic, usnic, bnxt_re, ocrdma, iser:
- remove unnecessary NULL checks
- kmap obsolescence
- pci_enable_pcie_error_reporting() obsolescence
- unused variables and macros
- trace event related warnings
- casting warnings
- Code cleanups for irdm and erdma
- EFA reporting of 128 byte PCIe TLP support
- mlx5 more agressively uses the out of order HW feature
- Big rework of how state machines and tasks work in rxe
- Fix a syzkaller found crash netdev refcount leak in siw
- bnxt_re revises their HW description header
- Congestion control for bnxt_re
- Use mmu_notifiers more safely in hfi1
- mlx5 gets better support for PCIe relaxed ordering inside VMs"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (81 commits)
RDMA/efa: Add rdma write capability to device caps
RDMA/mlx5: Use correct device num_ports when modify DC
RDMA/irdma: Drop spurious WQ_UNBOUND from alloc_ordered_workqueue() call
RDMA/rxe: Fix spinlock recursion deadlock on requester
RDMA/mlx5: Fix flow counter query via DEVX
RDMA/rxe: Protect QP state with qp->state_lock
RDMA/rxe: Move code to check if drained to subroutine
RDMA/rxe: Remove qp->req.state
RDMA/rxe: Remove qp->comp.state
RDMA/rxe: Remove qp->resp.state
RDMA/mlx5: Allow relaxed ordering read in VFs and VMs
net/mlx5: Update relaxed ordering read HCA capabilities
RDMA/mlx5: Check pcie_relaxed_ordering_enabled() in UMR
RDMA/mlx5: Remove pcie_relaxed_ordering_enabled() check for RO write
RDMA: Add ib_virt_dma_to_page()
RDMA/rxe: Fix the error "trying to register non-static key in rxe_cleanup_task"
RDMA/irdma: Slightly optimize irdma_form_ah_cm_frame()
RDMA/rxe: Fix incorrect TASKLET_STATE_SCHED check in rxe_task.c
IB/hfi1: Place struct mmu_rb_handler on cache line start
IB/hfi1: Fix bugs with non-PAGE_SIZE-end multi-iovec user SDMA requests
...
