Till recently it was not possible for userspace to specify a different
IOVA, but with the new ibv_reg_mr_iova() library call this can be done.
To compute the user_va we must compute:
user_va = (iova - iova_start) + user_va_start
while being cautious of overflow and other math problems.
The iova is not reliably stored in the mmkey when the MR is created. Only
the cached creation path (the common one) set it, so it must also be set
when creating uncached MRs.
Fix the weird use of iova when computing the starting page index in the
MR. In the normal case, when iova == umem.address:
iova & (~(BIT(page_shift) - 1)) ==
ALIGN_DOWN(umem.address, odp->page_size) ==
ib_umem_start(odp)
And when iova is different using it in math with a user_va is wrong.
Finally, do not allow an implicit ODP to be created with a non-zero IOVA
as we have no support for that.
Fixes: 7bdf65d411 ("IB/mlx5: Handle page faults")
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
The ODP handler for WQEs in RQ or SRQ is not implented for kernel QPs.
Therefore don't report support in these if query comes from a kernel user.
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
One of the steps in ODP page fault handler for WQEs is to read a WQE
from a QP send queue or receive queue buffer at a specific index.
Since the implementation of this buffer is different between kernel and
user QP the implementation of the handler needs to be aware of that and
handle it in a different way.
ODP for kernel MRs is currently supported only for RDMA_READ
and RDMA_WRITE operations so change the handler to
- read a WQE from a kernel QP send queue
- fail if access to receive queue or shared receive queue is
required for a kernel QP
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Add ib_reg_user_mr() for kernel ULPs to register user MRs.
The common use case that uses this function is a userspace application
that allocates memory for HCA access but the responsibility to register
the memory at the HCA is on an kernel ULP. This ULP that acts as an agent
for the userspace application.
This function is intended to be used without a user context so vendor
drivers need to be aware of calling reg_user_mr() device operation with
udata equal to NULL.
Among all drivers, i40iw is the only driver which relies on presence
of udata, so check udata existence for that driver.
Signed-off-by: Moni Shoua <monis@mellanox.com>
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
So far the assumption was that ib_umem_get() and ib_umem_odp_get()
are called from flows that start in UVERBS and therefore has a user
context. This assumption restricts flows that are initiated by ULPs
and need the service that ib_umem_get() provides.
This patch changes ib_umem_get() and ib_umem_odp_get() to get IB device
directly by relying on the fact that both UVERBS and ULPs sets that
field correctly.
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
With the new FD structure the async commands do not need to hold any
references while running. The existing mlx5_cmd_exec_cb() and
mlx5_cmd_cleanup_async_ctx() provide enough synchronization to ensure
that all outstanding commands are completed before the uobject can be
destructed.
Remove the now confusing get_file() and the type erasure of the
devx_async_cmd_event_file.
Link: https://lore.kernel.org/r/1578504126-9400-4-git-send-email-yishaih@mellanox.com
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
FD uobjects have a weird split between the struct file and uobject
world. Simplify this to make them pure uobjects and use a generic release
method for all struct file operations.
This fixes the control flow so that mlx5_cmd_cleanup_async_ctx() is always
called before erasing the linked list contents to make the concurrancy
simpler to understand.
For this to work the uobject destruction must fence anything that it is
cleaning up - the design must not rely on struct file lifetime.
Only deliver_event() relies on the struct file to when adding new events
to the queue, add a is_destroyed check under lock to block it.
Link: https://lore.kernel.org/r/1578504126-9400-3-git-send-email-yishaih@mellanox.com
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
dispatch_event_fd() runs from a notifier with minimal locking, and relies
on RCU and a file refcount to keep the uobject and eventfd alive.
As the next patch wants to remove the file_operations release function
from the drivers, re-organize things so that the devx_event_notifier()
path uses the existing RCU to manage the lifetime of the uobject and
eventfd.
Move the refcount puts to a call_rcu so that the objects are guaranteed to
exist and remove the indirect file refcount.
Link: https://lore.kernel.org/r/1578504126-9400-2-git-send-email-yishaih@mellanox.com
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When hardware is in resetting stage, we may can't poll back all the
expected work completions as the hardware won't generate cqe anymore.
This patch allows the driver to compose the expected wc instead of the
hardware during resetting stage. Once the hardware finished resetting, we
can poll cq from hardware again.
Link: https://lore.kernel.org/r/1578572412-25756-1-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
vm_ops is now initialized in ib_uverbs_mmap() with the recent rdma mmap
API changes. Earlier it was done in rdma_umap_priv_init() which would not
be called unless a driver called rdma_user_mmap_io() in its mmap.
i40iw does not use the rdma_user_mmap_io API but sets the vma's
vm_private_data to a driver object. This now conflicts with the vm_op
rdma_umap_close as priv pointer points to the i40iw driver object instead
of the private data setup by core when rdma_user_mmap_io is called. This
leads to a crash in rdma_umap_close with a mmap put being called when it
should not have.
Remove the redundant setting of the vma private_data in i40iw as it is not
used. Also move i40iw over to use the rdma_user_mmap_io API. This gives
the extra protection of having the mappings zapped when the context is
detsroyed.
BUG: unable to handle page fault for address: 0000000100000001
#PF: supervisor write access in kernel mode
#PF: error_code(0x0002) - not-present page
PGD 0 P4D 0
Oops: 0002 [#1] SMP PTI
CPU: 6 PID: 9528 Comm: rping Kdump: loaded Not tainted 5.5.0-rc4+ #117
Hardware name: Gigabyte Technology Co., Ltd. To be filled by O.E.M./Q87M-D2H, BIOS F7 01/17/2014
RIP: 0010:rdma_user_mmap_entry_put+0xa/0x30 [ib_core]
RSP: 0018:ffffb340c04c7c38 EFLAGS: 00010202
RAX: 00000000ffffffff RBX: ffff9308e7be2a00 RCX: 000000000000cec0
RDX: 0000000000000000 RSI: 0000000000000006 RDI: 0000000100000001
RBP: ffff9308dc7641f0 R08: 0000000000000001 R09: 0000000000000000
R10: 0000000000000001 R11: ffffffff8d4414d8 R12: ffff93075182c780
R13: 0000000000000001 R14: ffff93075182d2a8 R15: ffff9308e2ddc840
FS: 0000000000000000(0000) GS:ffff9308fdc00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000100000001 CR3: 00000002e0412004 CR4: 00000000001606e0
Call Trace:
rdma_umap_close+0x40/0x90 [ib_uverbs]
remove_vma+0x43/0x80
exit_mmap+0xfd/0x1b0
mmput+0x6e/0x130
do_exit+0x290/0xcc0
? get_signal+0x152/0xc40
do_group_exit+0x46/0xc0
get_signal+0x1bd/0xc40
? prepare_to_wait_event+0x97/0x190
do_signal+0x36/0x630
? remove_wait_queue+0x60/0x60
? __audit_syscall_exit+0x1d9/0x290
? rcu_read_lock_sched_held+0x52/0x90
? kfree+0x21c/0x2e0
exit_to_usermode_loop+0x4f/0xc3
do_syscall_64+0x1ed/0x270
entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x7fae715a81fd
Code: Bad RIP value.
RSP: 002b:00007fae6e163cb0 EFLAGS: 00000293 ORIG_RAX: 0000000000000001
RAX: fffffffffffffe00 RBX: 00007fae6e163d30 RCX: 00007fae715a81fd
RDX: 0000000000000010 RSI: 00007fae6e163cf0 RDI: 0000000000000003
RBP: 00000000013413a0 R08: 00007fae68000000 R09: 0000000000000017
R10: 0000000000000001 R11: 0000000000000293 R12: 00007fae680008c0
R13: 00007fae6e163cf0 R14: 00007fae717c9804 R15: 00007fae6e163ed0
CR2: 0000000100000001
---[ end trace b33d58d3a06782cb ]---
RIP: 0010:rdma_user_mmap_entry_put+0xa/0x30 [ib_core]
Fixes: b86deba977 ("RDMA/core: Move core content from ib_uverbs to ib_core")
Link: https://lore.kernel.org/r/20200107162223.1745-1-shiraz.saleem@intel.com
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
ioremap has provided non-cached semantics by default since the Linux 2.6
days, so remove the additional ioremap_nocache interface.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Building MR translation table in the ODP case requires additional
flexibility, namely random access to DMA addresses. Make both direct and
indirect ODP MR use same code path, separated from the non-ODP MR code
path.
With the restructuring the correct page_shift is now used around
__mlx5_ib_populate_pas().
Fixes: d2183c6f19 ("RDMA/umem: Move page_shift from ib_umem to ib_odp_umem")
Link: https://lore.kernel.org/r/20191222124649.52300-2-leon@kernel.org
Signed-off-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When a TID RDMA ACK to RESYNC request is received, the flow PSNs for
pending TID RDMA WRITE segments will be adjusted with the next flow
generation number, based on the resync_psn value extracted from the flow
PSN of the TID RDMA ACK packet. The resync_psn value indicates the last
flow PSN for which a TID RDMA WRITE DATA packet has been received by the
responder and the requester should resend TID RDMA WRITE DATA packets,
starting from the next flow PSN.
However, if resync_psn points to the last flow PSN for a segment and the
next segment flow PSN starts with a new generation number, use of the old
resync_psn to adjust the flow PSN for the next segment will lead to
miscalculation, resulting in WARN_ON and sge rewinding errors:
WARNING: CPU: 4 PID: 146961 at /nfs/site/home/phcvs2/gitrepo/ifs-all/components/Drivers/tmp/rpmbuild/BUILD/ifs-kernel-updates-3.10.0_957.el7.x86_64/hfi1/tid_rdma.c:4764 hfi1_rc_rcv_tid_rdma_ack+0x8f6/0xa90 [hfi1]
Modules linked in: ib_ipoib(OE) hfi1(OE) rdmavt(OE) rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfsv3 nfs_acl nfs lockd grace fscache iTCO_wdt iTCO_vendor_support skx_edac intel_powerclamp coretemp intel_rapl iosf_mbi kvm irqbypass crc32_pclmul ghash_clmulni_intel ib_isert iscsi_target_mod target_core_mod aesni_intel lrw gf128mul glue_helper ablk_helper cryptd rpcrdma sunrpc opa_vnic ast ttm ib_iser libiscsi drm_kms_helper scsi_transport_iscsi ipmi_ssif syscopyarea sysfillrect sysimgblt fb_sys_fops drm joydev ipmi_si pcspkr sg drm_panel_orientation_quirks ipmi_devintf lpc_ich i2c_i801 ipmi_msghandler wmi rdma_ucm ib_ucm ib_uverbs acpi_cpufreq acpi_power_meter ib_umad rdma_cm ib_cm iw_cm ip_tables ext4 mbcache jbd2 sd_mod crc_t10dif crct10dif_generic crct10dif_pclmul i2c_algo_bit crct10dif_common
crc32c_intel e1000e ib_core ahci libahci ptp libata pps_core nfit libnvdimm [last unloaded: rdmavt]
CPU: 4 PID: 146961 Comm: kworker/4:0H Kdump: loaded Tainted: G W OE ------------ 3.10.0-957.el7.x86_64 #1
Hardware name: Intel Corporation S2600WFT/S2600WFT, BIOS SE5C620.86B.0X.02.0117.040420182310 04/04/2018
Workqueue: hfi0_0 _hfi1_do_tid_send [hfi1]
Call Trace:
<IRQ> [<ffffffff9e361dc1>] dump_stack+0x19/0x1b
[<ffffffff9dc97648>] __warn+0xd8/0x100
[<ffffffff9dc9778d>] warn_slowpath_null+0x1d/0x20
[<ffffffffc05d28c6>] hfi1_rc_rcv_tid_rdma_ack+0x8f6/0xa90 [hfi1]
[<ffffffffc05c21cc>] hfi1_kdeth_eager_rcv+0x1dc/0x210 [hfi1]
[<ffffffffc05c23ef>] ? hfi1_kdeth_expected_rcv+0x1ef/0x210 [hfi1]
[<ffffffffc0574f15>] kdeth_process_eager+0x35/0x90 [hfi1]
[<ffffffffc0575b5a>] handle_receive_interrupt_nodma_rtail+0x17a/0x2b0 [hfi1]
[<ffffffffc056a623>] receive_context_interrupt+0x23/0x40 [hfi1]
[<ffffffff9dd4a294>] __handle_irq_event_percpu+0x44/0x1c0
[<ffffffff9dd4a442>] handle_irq_event_percpu+0x32/0x80
[<ffffffff9dd4a4cc>] handle_irq_event+0x3c/0x60
[<ffffffff9dd4d27f>] handle_edge_irq+0x7f/0x150
[<ffffffff9dc2e554>] handle_irq+0xe4/0x1a0
[<ffffffff9e3795dd>] do_IRQ+0x4d/0xf0
[<ffffffff9e36b362>] common_interrupt+0x162/0x162
<EOI> [<ffffffff9dfa0f79>] ? swiotlb_map_page+0x49/0x150
[<ffffffffc05c2ed1>] hfi1_verbs_send_dma+0x291/0xb70 [hfi1]
[<ffffffffc05c2c40>] ? hfi1_wait_kmem+0xf0/0xf0 [hfi1]
[<ffffffffc05c3f26>] hfi1_verbs_send+0x126/0x2b0 [hfi1]
[<ffffffffc05ce683>] _hfi1_do_tid_send+0x1d3/0x320 [hfi1]
[<ffffffff9dcb9d4f>] process_one_work+0x17f/0x440
[<ffffffff9dcbade6>] worker_thread+0x126/0x3c0
[<ffffffff9dcbacc0>] ? manage_workers.isra.25+0x2a0/0x2a0
[<ffffffff9dcc1c31>] kthread+0xd1/0xe0
[<ffffffff9dcc1b60>] ? insert_kthread_work+0x40/0x40
[<ffffffff9e374c1d>] ret_from_fork_nospec_begin+0x7/0x21
[<ffffffff9dcc1b60>] ? insert_kthread_work+0x40/0x40
This patch fixes the issue by adjusting the resync_psn first if the flow
generation has been advanced for a pending segment.
Fixes: 9e93e967f7 ("IB/hfi1: Add a function to receive TID RDMA ACK packet")
Link: https://lore.kernel.org/r/20191219231920.51069.37147.stgit@awfm-01.aw.intel.com
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>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The current debugfs output for receive contexts (rcds), stops after the
kernel receive contexts have been displayed. This is not enough
information to fully diagnose packet drops.
Display all of the receive contexts.
Augment the output with some more context information.
Limit the ring buffer header output to 5 entries to avoid overextending
the sequential file output.
Fixes: bf808b5039 ("IB/hfi1: Add kernel receive context info to debugfs")
Link: https://lore.kernel.org/r/20191219211928.58387.20737.stgit@awfm-01.aw.intel.com
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
In the iowait structure, two iowait_work entries were included to queue a
given object: one for normal IB operations, and the other for TID RDMA
operations. For non-TID RDMA operations, the iowait_work structure for TID
RDMA is initialized to contain a NULL function (not used). When the QP is
reset, the function iowait_cancel_work will be called to cancel any
pending work. The problem is that this function will call
cancel_work_sync() for both iowait_work entries, even though the one for
TID RDMA is not used at all. Eventually, the call cascades to
__flush_work(), wherein a WARN_ON will be triggered due to the fact that
work->func is NULL.
The WARN_ON was introduced in commit 4d43d395fe ("workqueue: Try to
catch flush_work() without INIT_WORK().")
This patch fixes the issue by making sure that a work function is present
for TID RDMA before calling cancel_work_sync in iowait_cancel_work.
Fixes: 4d43d395fe ("workqueue: Try to catch flush_work() without INIT_WORK().")
Fixes: 5da0fc9dbf ("IB/hfi1: Prepare resource waits for dual leg")
Link: https://lore.kernel.org/r/20191219211941.58387.39883.stgit@awfm-01.aw.intel.com
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>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
