This patch adds the counter n_tidwait to count the number of times the
TID resource allocator has to wait for TID resources.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch adds the OPFN helper functions to initialize, encode, decode,
and reset OPFN parameters for the TID RDMA feature.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This has been a fairly typical cycle, with the usual sorts of driver
updates. Several series continue to come through which improve and
modernize various parts of the core code, and we finally are starting to
get the uAPI command interface cleaned up.
- Various driver fixes for bnxt_re, cxgb3/4, hfi1, hns, i40iw, mlx4, mlx5,
qib, rxe, usnic
- Rework the entire syscall flow for uverbs to be able to run over
ioctl(). Finally getting past the historic bad choice to use write()
for command execution
- More functional coverage with the mlx5 'devx' user API
- Start of the HFI1 series for 'TID RDMA'
- SRQ support in the hns driver
- Support for new IBTA defined 2x lane widths
- A big series to consolidate all the driver function pointers into
a big struct and have drivers provide a 'static const' version of the
struct instead of open coding initialization
- New 'advise_mr' uAPI to control device caching/loading of page tables
- Support for inline data in SRPT
- Modernize how umad uses the driver core and creates cdev's and sysfs
files
- First steps toward removing 'uobject' from the view of the drivers
-----BEGIN PGP SIGNATURE-----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=T0p1
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"This has been a fairly typical cycle, with the usual sorts of driver
updates. Several series continue to come through which improve and
modernize various parts of the core code, and we finally are starting
to get the uAPI command interface cleaned up.
- Various driver fixes for bnxt_re, cxgb3/4, hfi1, hns, i40iw, mlx4,
mlx5, qib, rxe, usnic
- Rework the entire syscall flow for uverbs to be able to run over
ioctl(). Finally getting past the historic bad choice to use
write() for command execution
- More functional coverage with the mlx5 'devx' user API
- Start of the HFI1 series for 'TID RDMA'
- SRQ support in the hns driver
- Support for new IBTA defined 2x lane widths
- A big series to consolidate all the driver function pointers into a
big struct and have drivers provide a 'static const' version of the
struct instead of open coding initialization
- New 'advise_mr' uAPI to control device caching/loading of page
tables
- Support for inline data in SRPT
- Modernize how umad uses the driver core and creates cdev's and
sysfs files
- First steps toward removing 'uobject' from the view of the drivers"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (193 commits)
RDMA/srpt: Use kmem_cache_free() instead of kfree()
RDMA/mlx5: Signedness bug in UVERBS_HANDLER()
IB/uverbs: Signedness bug in UVERBS_HANDLER()
IB/mlx5: Allocate the per-port Q counter shared when DEVX is supported
IB/umad: Start using dev_groups of class
IB/umad: Use class_groups and let core create class file
IB/umad: Refactor code to use cdev_device_add()
IB/umad: Avoid destroying device while it is accessed
IB/umad: Simplify and avoid dynamic allocation of class
IB/mlx5: Fix wrong error unwind
IB/mlx4: Remove set but not used variable 'pd'
RDMA/iwcm: Don't copy past the end of dev_name() string
IB/mlx5: Fix long EEH recover time with NVMe offloads
IB/mlx5: Simplify netdev unbinding
IB/core: Move query port to ioctl
RDMA/nldev: Expose port_cap_flags2
IB/core: uverbs copy to struct or zero helper
IB/rxe: Reuse code which sets port state
IB/rxe: Make counters thread safe
IB/mlx5: Use the correct commands for UMEM and UCTX allocation
...
When it is requested to change its physical state back to Offline while in
the process to go up, DC8051 will set the ERROR field in the
DC8051_DBG_ERR_INFO_SET_BY_8051 register. This ERROR field will remain
until the next time when DC8051 transitions from Offline to Polling.
Subsequently, when the host requests DC8051 to change its physical state
to Polling again, it may receive a DC8051 interrupt with the stale ERROR
field still in DC8051_DBG_ERR_INFO_SET_BY_8051. If the host link state has
been changed to Polling, this stale ERROR will force the host to
transition to Offline state, resulting in a vicious cycle of Polling
->Offline->Polling->Offline. On the other hand, if the host link state is
still Offline when the stale ERROR is received, the stale ERROR will be
ignored, and the link will come up correctly. This patch implements the
correct behavior by changing host link state to Polling only after DC8051
changes its physical state to Polling.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Krzysztof Goreczny <krzysztof.goreczny@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>
From git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git
This is required to resolve dependencies of the next series of RDMA
patches.
The code motion conflicts in drivers/infiniband/core/cache.c were
resolved.
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Since Virtual Lanes BCT credits and MTU are set through separate MADs, we
have to ensure both are valid, and data VLs are ready for transmission
before we allow port transition to Armed state.
Fixes: 5e2d6764a7 ("IB/hfi1: Verify port data VLs credits on transition to Armed")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
rvt_destroy_qp() cannot complete until all in process packets have
been released from the underlying hardware. If a link down event
occurs, an application can hang with a kernel stack similar to:
cat /proc/<app PID>/stack
quiesce_qp+0x178/0x250 [hfi1]
rvt_reset_qp+0x23d/0x400 [rdmavt]
rvt_destroy_qp+0x69/0x210 [rdmavt]
ib_destroy_qp+0xba/0x1c0 [ib_core]
nvme_rdma_destroy_queue_ib+0x46/0x80 [nvme_rdma]
nvme_rdma_free_queue+0x3c/0xd0 [nvme_rdma]
nvme_rdma_destroy_io_queues+0x88/0xd0 [nvme_rdma]
nvme_rdma_error_recovery_work+0x52/0xf0 [nvme_rdma]
process_one_work+0x17a/0x440
worker_thread+0x126/0x3c0
kthread+0xcf/0xe0
ret_from_fork+0x58/0x90
0xffffffffffffffff
quiesce_qp() waits until all outstanding packets have been freed.
This wait should be momentary. During a link down event, the cleanup
handling does not ensure that all packets caught by the link down are
flushed properly.
This is caused by the fact that the freeze path and the link down
event is handled the same. This is not correct. The freeze path
waits until the HFI is unfrozen and then restarts PIO. A link down
is not a freeze event. The link down path cannot restart the PIO
until link is restored. If the PIO path is restarted before the link
comes up, the application (QP) using the PIO path will hang (until
link is restored).
Fix by separating the linkdown path from the freeze path and use the
link down path for link down events.
Close a race condition sc_disable() by acquiring both the progress
and release locks.
Close a race condition in sc_stop() by moving the setting of the flag
bits under the alloc lock.
Cc: <stable@vger.kernel.org> # 4.9.x+
Fixes: 7724105686 ("IB/hfi1: add driver files")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
HFI IRQ enable bits are not being set correctly. Send context error and
DC IRQs are not being enabled correctly. In addition, send context error
IRQs are not being delivered.
Because of this, send context errors are not being handled correctly when
they occur.
When setting the IRQ bits, if an IRQ range is used, and the last bit is on
a register boundary (bit 63), the calculated index for the final register
modification is incorrect (index + 1 vs. index).
The incorrect index calculation causes incorrect IRQ bits to be set. In
this case the send context error IRQ is NOT enabled.
Fix by using the 'last' value rather than the counted 'src' value to
determine the final index to use. This satisfies all cases.
Fixes: a2f7bbdc2d ("IB/hfi1: Rework the IRQ API to be more flexible")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
User contexts use the receive URGENT interrupt. However, enabling
the IRQ SRC in the file_ops module is not as clean as it could be.
Augment the _rcvctl() function to be able to enable/disable the IRQ
source.
Use the new interface from file_ops to enable/disable the IRQ.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current IRQ API is an all or nothing interface. This has two
problems:
1. All IRQs are enabled regardless of use
2. Moving from general interrupt to MSIx handling is difficult
Introduce a new API to enable/disable specific IRQs or a range of IRQs.
Do not enable and disable all IRQs in one step.
Rework various modules to enable/disable IRQs when needed.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current method of allocating MSIx resources is a bit cumbersome,
and not very easily added to.
Refactor and re-order the code paths into a more consistent interface.
Update the interface so that allocations are not order dependent.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current HFI1 MSIx API is difficult to follow, change, or add to.
In anticipation of moving to an more flexible API, move the current
MSIx functionality to the new msix.c module.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Currently several things occur before the hfi1_devdata structure is
allocated. This leads to an inconsistent logging ability and makes
it more difficult to restructure some code paths.
Allocate (and do a minimal init) the structure as soon as possible.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The tune_pcie_caps needs to occur sometime after PCI is enabled, but
before the HFI is enabled. Currently it is placed in the MSIx
allocation code which doesn't really fit. Moving it to just after
the gen3 bump.
Clean up the associated code (modules, etc.).
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The general interrupt handler is_rcv_avail_int() has two paths,
do_interrupt() (callback) and handle_user_interrupt(). The
do_interrupt() callback is for the threaded receive handling.
is_rcv_avail_int() cannot handle threaded IRQs.
If the do_interrupt() path is taken, and the IRQ returns
IRQ_WAKE_THREAD, the IRQ behavior will be indeterminate.
Remove incorrect call to do_interrupt() from is_rcv_avail_int(),
leaving the un-threaded (handle_user_interrupt()) path.
Fixes: f4f30031c3 ("staging/rdma/hfi1: Thread the receive interrupt.")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The in_use_ctxts bitmask is for user receive contexts only. Setting it for
any other type of receive context is incorrect.
Move initial set of in_use_ctxts bits from the general context init to the
user context specific init. Having this bit set can allow contexts to be
incorrectly identified by some IRQ handlers. This will allow
handle_user_interrupt() will now filter user contexts correctly.
Clean up redundant is_rcv_urgent_int() user context check.
A follow on patch will clean up an incorrect code path in the
is_rcv_avail_int().
Fixes: 8737ce95c4 ("IB/hfi1: Fix an assign/ordering issue with shared context IDs")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The INTx IRQ support does not work for all HF1 IRQ handlers
(specifically the receive data IRQs).
Remove all supporting code for the INTx IRQ.
If the requested MSIx vector request is unsuccessful, do not allow the
driver to continue.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Remove the sizeable cache of the chip sizing CSRs and replace with CSR
reads as needed.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
It is only ever written.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The usage of this ctxt data field is not hot path and the value can be
computed on demand to cut down the ctxtdata bloat.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The field is based on a constant that can never change.
Use the define to assign the register instead.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This field should be in ctxtdata to allow for better locality of access by
eliminating a dd dereference.
The new field is now side-by-side with rcvhdrqentsize since the rhf_offset
is a function of the rcvhdrqentsize.
Both fields are now correctly sized as u8.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
These registers were not added in the 16B work.
Add them and replace blind constants with the correct defines.
Fixes: 72c07e2b67 ("IB/hfi1: Add support to receive 16B bypass packets")
Reviewed-by: Don Hiatt <don.hiatt@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The following code fails to allocate a buffer for the
tail address that the hardware DMAs into when the user
context DMA_RTAIL is set.
if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
rcd->rcvhdrtail_kvaddr = dma_zalloc_coherent(
&dd->pcidev->dev, PAGE_SIZE, &dma_hdrqtail,
gfp_flags);
if (!rcd->rcvhdrtail_kvaddr)
goto bail_free;
rcd->rcvhdrqtailaddr_dma = dma_hdrqtail;
}
So the rcvhdrtail_kvaddr would then be NULL.
The mmap logic fails to check for a NULL rcvhdrtail_kvaddr.
The fix is to test for both user and kernel DMA_TAIL options
during the allocation as well as testing for a NULL
rcvhdrtail_kvaddr during the mmap processing.
Additionally, all downstream testing of the capmask for DMA_RTAIL
have been eliminated in favor of testing rcvhdrtail_kvaddr.
Cc: <stable@vger.kernel.org> # 4.9.x
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Update mlx4 to support user MR creation against read-only memory, previously
it required the memory to be writable.
Based on rdma for-rc due to dependencies.
* mr_fix: (2 commits)
IB/mlx4: Mark user MR as writable if actual virtual memory is writable
IB/core: Make testing MR flags for writability a static inline function
Currently the driver doesn't support completion vectors. These
are used to indicate which sets of CQs should be grouped together
into the same vector. A vector is a CQ processing thread that
runs on a specific CPU.
If an application has several CQs bound to different completion
vectors, and each completion vector runs on different CPUs, then
the completion queue workload is balanced. This helps scale as more
nodes are used.
Implement CQ completion vector support using a global workqueue
where a CQ entry is queued to the CPU corresponding to the CQ's
completion vector. Since the workqueue is global, it's guaranteed
to always be there when queueing CQ entries; Therefore, the RCU
locking for cq->rdi->worker in the hot path is superfluous.
Each completion vector is assigned to a different CPU. The number of
completion vectors available is computed by taking the number of
online, physical CPUs from the local NUMA node and subtracting the
CPUs used for kernel receive queues and the general interrupt.
Special use cases:
* If there are no CPUs left for completion vectors, the same CPU
for the general interrupt is used; Therefore, there would only
be one completion vector available.
* For multi-HFI systems, the number of completion vectors available
for each device is the total number of completion vectors in
the local NUMA node divided by the number of devices in the same
NUMA node. If there's a division remainder, the first device to
get initialized gets an extra completion vector.
Upon a CQ creation, an invalid completion vector could be specified.
Handle it as follows:
* If the completion vector is less than 0, set it to 0.
* Set the completion vector to the result of the passed completion
vector moded with the number of device completion vectors
available.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When Hfi1 device is unresponsive, reading the RcvArrayCnt register
will return all 1's. This value is then used to remap chip's RcvArray.
The incorrect all ones value used in remapping RcvArray
will cause warn on as shown by trace below:
[<ffffffff81685eac>] dump_stack+0x19/0x1b
[<ffffffff81085820>] warn_slowpath_common+0x70/0xb0
[<ffffffff810858bc>] warn_slowpath_fmt+0x5c/0x80
[<ffffffff81065c29>] __ioremap_caller+0x279/0x320
[<ffffffff8142873c>] ? _dev_info+0x6c/0x90
[<ffffffffa021d155>] ? hfi1_pcie_ddinit+0x1d5/0x330 [hfi1]
[<ffffffff81065d62>] ioremap_wc+0x32/0x40
[<ffffffffa021d155>] hfi1_pcie_ddinit+0x1d5/0x330 [hfi1]
[<ffffffffa0204851>] hfi1_init_dd+0x1d1/0x2440 [hfi1]
[<ffffffff813503dc>] ? pci_write_config_word+0x1c/0x20
Read CCE revision register first to verify that WFR device is
responsive. If the read return "all ones", bail out from init
and fail the driver load.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The packet fault injection code present in the HFI1 driver had some
issues which not only fragment the code but also created user
confusion. Furthermore, it suffered from the following issues:
1. The fault_packet method only worked for received packets. This
meant that the only fault injection mode available for sent
packets is fault_opcode, which did not allow for random packet
drops on all egressing packets.
2. The mask available for the fault_opcode mode did not really work
due to the fact that the opcode values are not bits in a bitmask but
rather sequential integer values. Creating a opcode/mask pair that
would successfully capture a set of packets was nearly impossible.
3. The code was fragmented and used too many debugfs entries to
operate and control. This was confusing to users.
4. It did not allow filtering fault injection on a per direction basis -
egress vs. ingress.
In order to improve or fix the above issues, the following changes have
been made:
1. The fault injection methods have been combined into a single fault
injection facility. As such, the fault injection has been plugged
into both the send and receive code paths. Regardless of method used
the fault injection will operate on both egress and ingress packets.
2. The type of fault injection - by packet or by opcode - is now controlled
by changing the boolean value of the file "opcode_mode". When the value
is set to True, fault injection is done by opcode. Otherwise, by
packet.
2. The masking ability has been removed in favor of a bitmap that holds
opcodes of interest (one bit per opcode, a total of 256 bits). This
works in tandem with the "opcode_mode" value. When the value of
"opcode_mode" is False, this bitmap is ignored. When the value is
True, the bitmap lists all opcodes to be considered for fault injection.
By default, the bitmap is empty. When the user wants to filter by opcode,
the user sets the corresponding bit in the bitmap by echo'ing the bit
position into the 'opcodes' file. This gets around the issue that the set
of opcodes does not lend itself to effective masks and allow for extremely
fine-grained filtering by opcode.
4. fault_packet and fault_opcode methods have been combined. Hence, there
is only one debugfs directory controlling the entire operation of the
fault injection machinery. This reduces the number of debugfs entries
and provides a more unified user experience.
5. A new control files - "direction" - is provided to allow the user to
control the direction of packets, which are subject to fault injection.
6. A new control file - "skip_usec" - is added that would allow the user
to specify a "timeout" during which no fault injection will occur.
In addition, the following bug fixes have been applied:
1. The fault injection code has been split into its own header and source
files. This was done to better organize the code and support conditional
compilation without littering the code with #ifdef's.
2. The method by which the TX PIO packets were being marked for drop
conflicted with the way send contexts were being setup. As a result,
the send context was repeatedly being reset.
3. The fault injection only makes sense when the user can control it
through the debugfs entries. However, a kernel configuration can
enable fault injection but keep fault injection debugfs entries
disabled. Therefore, it makes sense that the HFI fault injection
code depends on both.
4. Error suppression did not take into account the method by which PIO
packets were being dropped. Therefore, even with error suppression
turned on, errors would still be displayed to the screen. A larger
enough packet drop percentage would case the kernel to crash because
the driver would be stuck printing errors.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Don Hiatt <don.hiatt@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The driver_pstate() function is used to map internal driver state
information to externally defined states.
The VERIFY_CAP and GOING_UP states are config/training states, but
the mapping routing returns the POLLING value.
Update the return values for VERIFY_CAP and GOING_UP to return the
correct value: TRAINING.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When the LCB isn't able to get any lanes operational on the
first transition into mission mode, the link transfer active
never happens and the LNI stays in the polling state indefinitely.
Reset LCB upon receiving an 8051 interrupt for LCB to try to obtain
lanes with firmware version 1.25.0 or later. Also, update the LCB
reset value in other parts of the code with a macro defined to make
the code more maintainable and rename functions with the link_width
label to link_mode to reflect the fact that those functions set and
read link related data not just the link width.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A pio send egress error can occur when the PSM library attempts to
to send a bad packet. That issue is still being investigated.
The pio error interrupt handler then attempts to progress the recovery
of the errored pio send context.
Code inspection reveals that the handling lacks the necessary locking
if that recovery interleaves with a PSM close of the "context" object
contains the pio send context.
The lack of the locking can cause the recovery to access the already
freed pio send context object and incorrectly deduce that the pio
send context is actually a kernel pio send context as shown by the
NULL deref stack below:
[<ffffffff8143d78c>] _dev_info+0x6c/0x90
[<ffffffffc0613230>] sc_restart+0x70/0x1f0 [hfi1]
[<ffffffff816ab124>] ? __schedule+0x424/0x9b0
[<ffffffffc06133c5>] sc_halted+0x15/0x20 [hfi1]
[<ffffffff810aa3ba>] process_one_work+0x17a/0x440
[<ffffffff810ab086>] worker_thread+0x126/0x3c0
[<ffffffff810aaf60>] ? manage_workers.isra.24+0x2a0/0x2a0
[<ffffffff810b252f>] kthread+0xcf/0xe0
[<ffffffff810b2460>] ? insert_kthread_work+0x40/0x40
[<ffffffff816b8798>] ret_from_fork+0x58/0x90
[<ffffffff810b2460>] ? insert_kthread_work+0x40/0x40
This is the best case scenario and other scenarios can corrupt the
already freed memory.
Fix by adding the necessary locking in the pio send context error
handler.
Cc: <stable@vger.kernel.org> # 4.9.x
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
HFI's counters SendWaitCnt and SendWaitVlCnt are in units
of TXE cycle time (at 805MHz). OPA counters PortXmitWait and
PortVLXmtWait are in units of flit times.
Convert the counter values to flit units using following
conversion formula:
PortXmitWait =
SendWaitCnt * 2 * (4 /link_width) * (25 Gbps /link_speed)
PortVLXmitWait =
SendWaitVLCnt * 2 * (4 /link_width) * (25 Gbps /link_speed)
At link up or downgrade events, the link width can change. To ensure
accurate counter calculations, sample the counters after the events,
during counter requests, and then aggregate the OPA counters.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The pci_request_irq() interfaces always adds the IRQF_SHARED bit to
all IRQ requests.
When the kernel is built with CONFIG_DEBUG_SHIRQ config flag, if the
IRQF_SHARED bit is set, a call to the IRQ handler is made from the
__free_irq() function. This is testing a race condition between the
IRQ cleanup and an IRQ racing the cleanup. The HFI driver should be
able to handle this race, but does not.
This race can cause traces that start with this footprint:
BUG: unable to handle kernel NULL pointer dereference at (null)
Call Trace:
<hfi1 irq handler>
...
__free_irq+0x1b3/0x2d0
free_irq+0x35/0x70
pci_free_irq+0x1c/0x30
clean_up_interrupts+0x53/0xf0 [hfi1]
hfi1_start_cleanup+0x122/0x190 [hfi1]
postinit_cleanup+0x1d/0x280 [hfi1]
remove_one+0x233/0x250 [hfi1]
pci_device_remove+0x39/0xc0
Export IRQ cleanup function so it can be called from other modules.
Using the exported cleanup function:
Re-order the driver cleanup code to clean up IRQ resources before
other resources, eliminating the race.
Re-order error path for init so that the race does not occur.
Reduce severity on spurious error message for SDMA IRQs to info.
Reviewed-by: Alex Estrin <alex.estrin@intel.com>
Reviewed-by: Patel Jay P <jay.p.patel@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The get_unit_name() function crafts a string based on the device name
and the device unit number. It then stores this in a static variable.
This has concurrency issues as can be seen with this log:
hfi1 0000:02:00.0: hfi1_1: read_idle_message: read idle message 0x203
hfi1 0000:01:00.0: hfi1_1: read_idle_message: read idle message 0x203
The PCI device ID (0000:02:00.0 vs. 0000:01:00.0) is correct for the
message, but the device string hfi1_1 is incorrect (it should be
hfi1_0 for the second log message).
Remove get_unit_name() function.
Instead, use the rvt accessor rvt_get_ibdev_name() to get the IB name
string.
Clean up any hfi1_early_xx calls that can now use the new path.
QIB has the same (qib_get_unit_name()) issue. Updating as necessary.
Remove qib_get_unit_name() function.
Update log message that has redundant device name.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When an 8051 command times out, the entire DC block is restarted. During
the restart, the host interface version bit is set, which calls
do_8051_command() recursively. The host version bit needs to be set
before the link moves into polling, so the host version bit can be set
in set_local_link_attributes() instead. Thus, the 8051 command functions
can be simplied as a non-locking version (dd->dc8051_lock) of those
functions are no longer needed.
Fixes: 9be6a5d788 ("IB/hfi1: Prevent LNI out of sync by resetting host interface version")
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
On host shutdown, driver sends 'SMA_Disabled' as a reason
for link down. This is incorrect.
Send 'reboot' as a linkdown reason.
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
OPA VNIC does not use PIO contexts and instead only uses SDMA
engines. Do not allocate PIO contexts for VNIC ports.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
HFI's are hard-wired to send Device Info frames with
MgmtAllowed bit set to 0. This means in B2B setups,
MgmtAllowed would never be allowed, which prevents
remote opa management tools from working properly.
Assume MgmtAllowed if a neighbor is also an HFI.
Fixes: 98b9ee2002 ("IB/hfi1: Cache neighbor secure data after link up")
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The driver parameter num_user_contexts controls global behavior and
should not be modified by the driver.
This patch eliminates modification of num_user_contexts by using a
local variable to keep track of the value.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The only use for the urg interrupt is for priority PSM packets.
There is no reason for this interrupt to be enabled for kernel
contexts.
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
__subn_get_opa_portinfo stores value returned by hfi1_get_ib_cfg() as
operational vls. hfi1_get_ib_cfg() returns vls_operational field in
hfi1_pportdata. The problem with this is that the value is always equal
to vls_supported field in hfi1_pportdata.
The logic to calculate operational_vls is to set value passed by FM
(in __subn_set_opa_portinfo routine). If no value is passed then
default value is stored in operational_vls.
Field actual_vls_operational is calculated on the basis of buffer
control table. Hence, modifying hfi1_get_ib_cfg() to return
actual_operational_vls when used with HFI1_IB_CFG_OP_VLS parameter
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Patel Jay P <jay.p.patel@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The handler for link init state (HLS_UP_INIT) notifies userspace
(update_statusp()) before enabling the device
(RCV_CTRL_RCV_PORT_ENABLE_SMASK) or setting the device state
(ppd->host_link_state). This causes a race condition where the
userspace thinks the interface is in the INIT state before the driver
has set that state.
Rework the code path to eliminate the race.
Delay setting the init state until after a HW settling period.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Conflicts:
drivers/infiniband/hw/cxgb4/cm.c
drivers/infiniband/hw/qib/qib_driver.c
drivers/infiniband/hw/qib/qib_mad.c
There were minor fixups needed in these files. Just minor context diffs
due to patches from independent sources touching the same basic area.
Signed-off-by: Doug Ledford <dledford@redhat.com>
The early for-next branch was based on v4.14-rc2, while the shared pull
request I got from Mellanox used a v4.14-rc4 base. I'm making the
branch that was the shared Mellanox pull request the new for-next branch
and merging the early for-next branch into it.
Signed-off-by: Doug Ledford <dledford@redhat.com>
In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly. Switches test of .data field to
.function, since .data will be going away.
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Change serdes mode setting to use MISC_CONFIG_BITS in
VERIFY_CAP_LOCAL_LINK_WIDTH register. This method of
setting up serdes loopback is universally compatible
across all firmware versions.
Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com>
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The s_ahgpsn was incorrectly placed in the read-mostly section of the QP
and the s_curr_size and s_hdrwords are oversized. The misplaced
s_ahgpsn will cause the read-mostly cachelines to thrash.
Place s_ahgpsn in the send side cache lines and correctly size and
s_hdrwords and s_cur_size to keep the send side cachelines at the same
size.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When the link is disabled and re-enabled, the host version bit is not
set again, so the firmware behaves as though it’s interacting with an
old driver. This causes LNI to get out of sync. The host version bit
needs to be set at load_8051_firmware() and _dc_start(). Currently, it's
only set at load_8051_firmware().
Create a common function to set the bit with the intent to make the code
more maintainable in the future, set the host version bit at _dc_start()
and modify the 8051 command API to prevent a deadlock as _dc_start() is
already holding the dc8051 lock.
Fixes: 913cc67159 ("IB/hfi1: Always perform offline transition")
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
