Currently each user context is assigned a single SDMA engine
based on the VL, context id, and subcontext id. That means for
MPI applications, each rank can only use one SDMA engine for
all messages. This may create unwanted backup for independent
messages going to different destinations upon congestion at one
destination.
This patch adds the packet "dlid" to the formula of SDMA engine
selection for user SDMA requests. A simple hash table is used
to maintain even distribution among the available SDMA engines
regardless how the "dlid" values are distributed.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Remove the TWSI code. The driver now uses the kernel's built-in
i2c bit bus module.
Cc: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Use built-in i2c bit-shift bus adapter to control the
i2c busses on the chip.
Cc: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When performing process affinity recommendations for MPI ranks, the current
algorithm doesn't take into account multiple HFI units. Also, real
cores and HT cores are not distinguished from one another. Therefore,
all HT cores are recommended to be assigned first within the local NUMA
node before recommending the assignments of cores in other NUMA nodes.
It's ideal to assign all real cores across all NUMA nodes first, then all
HT 1 cores, then all HT 2 cores, and so on to balance CPU workload. CPU
cores in other NUMA nodes could be running interrupt handlers, and this is
not taken into account.
To balance the CPU workload for user processes, the following
recommendation algorithm is used:
For each user process that is opening a context on HFI Y:
a) If all cores are assigned to user processes, start assignments all
over from the first core
b) Assign real cores first, then HT cores (First set of HT cores on
all physical cores, then second set of HT cores, and, so on) in the
following order:
1. Same NUMA node as HFI Y and not running an IRQ handler
2. Same NUMA node as HFI Y and running an IRQ handler
3. Different NUMA node to HFI Y and not running an IRQ handler
4. Different NUMA node to HFI Y and running an IRQ handler
c) Mark core as assigned in the global affinity structure. As user
processes are done, remove core assignments from global affinity
structure.
This implementation allows an arbitrary number of HT cores and provides
support for multiple HFIs.
This is being included in the kernel rather than user space due to the
fact that user space has no way of knowing the CPU recommendations for
contexts running as part of other jobs.
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Kernel receive queues oversubscribe CPU cores on multi-HFI systems.
To prevent this, the kernel receive queues are separated onto
different cores, and the SDMA engine interrupts are constrained to
a lesser number of cores.
hfi1s_on_numa_node*krcvqs is the number of CPU cores that are
reserved for kernel receive queues for all HFIs. Each HFI initializes
its kernel receive queues to one of the reserved CPU cores. If there
ends up being 0 CPU cores leftover for SDMA engines, use the same
CPU cores as receive contexts.
In addition, general and control contexts are assigned to their own
CPU core, however, both types of contexts tend to have low traffic.
To save CPU cores, collapse general and control contexts to one CPU
core for all HFI units. This change prevents SDMA engine interrupts
from wrapping around general contexts.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When HFI units get initialized, they each use their own mask copy for
affinity assignments. On a multi-HFI system, affinity assignments
overbook CPU cores as each HFI doesn't have knowledge of affinity
assignments for other HFI units. Therefore, some CPU cores are never
used for interrupt handlers in systems with high number of CPU cores
per NUMA node.
For multi-HFI systems, SDMA engine interrupt assignments start all over
from the first CPU in the local NUMA node after the first HFI
initialization. This change allows assignments to continue where the
last HFI unit left off.
Add global structure for affinity assignments for multiple HFIs to share
affinity mask.
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Reviewed-by: Jubin John <jubin.john@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add sw counter to track dropped unsupported packets.
Report unsupported packets drop as the RcvError.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jakub Pawlak <jakub.pawlak@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add per VL XmitDiscards counters to the opapmaquery
status and error response.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Jakub Pawlak <jakub.pawlak@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Fix sparse errors by making sure the fast assign destinations
are host cpu typed.
For the void __iomem *, just make the field match source
data.
Fix a bug where the hw_free trace printed the pointer vs.
the dereferenced value.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@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>
The ftrace infrastructure used to evaluate the TRACE_SYSTEM
macro on every DEFINE_EVENT() macro. Now the TRACE_SYSTEM
macro only gets evaluated when trace/define_trace.h is
included, so the group event information is lost. This was
introduced in
commit acd388fd3a ("tracing: Give system name a pointer")
Therefore, each system tracepoint must be on its own file.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@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>
Fix a copy and paste typo in comment.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Simple code clean up of hfi1_write_iter.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The definition of port state changed mid development and the
old structure was kept accidentally. Remove this dead code.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The critical section should protect only the list traversal
and dd->asic_data modification, not the memory allocation.
The fix pulls the allocation out of the critical section.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There are several computatations of the sc in the
ud receive routine.
Besides the code duplication, all are wrong when the
sc is greater than 15. In that case the code incorrectly
or's a 1 into the computed sc instead of 1 shifted left
by 4.
Fix precomputed sc5 by using an already implemented routine
hdr2sc() and deleting flawed duplicated code.
Cc: Stable <stable@vger.kernel.org> # 4.6+
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@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>
A failure in the get_txreq() inline will result in a
slow path retry using __get_txreq().
__get_txreq() attempts to procure the qp s_lock, which
is already held in all callers.
Fix by deleting the s_lock maintenance in __get_txreq()
and add sparse syntax hooks to future proof the code.
Cc: Stable <stable@vger.kernel.org> # 4.6+
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@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>
Swapping a cable from a "Mgmt Allowed=No" switch port to a
"Mgmt Allowed=Yes" switch port doesn't send a pkey change
notification. Therefore, the link doesn't become active as
the oib_utils layer uses an old pkey table cache.
Fix by ensuring the pkey change notification is sent when
the table is changed both explicitly by the FM and implicitly
by the driver via a cable swap.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
FULL_MGMT_P_KEY doesn't get cleared from the pkey table at link bounce
because the link down and link bounce code paths are different when
moving a QSFP cable on a switch. This causes an HFI unit connected to a
switch to try to be initialized to an FM node when the QSFP cable is
moved from a MgmtAllowed=NO port to a MgmtAllowed=YES port and back to a
MgmtAllowed=NO port. Remove FULL_MGMT_P_KEY from pkey table at link up.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This fixes potential buffer overflow because the sprintf function
doesn't check buffer boundaries. Use snprintf instead.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This fixes potential NULL ptr dereference because IS_ERR(dd) doesn't
handle NULL. Fix the issue by initializing the pointer with a not NULL
error code.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
If a context has already been assigned to an FD, prevent
another assignment.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current value of 500us for the packet egress timeout is too small
which causes the host to declare failure on draining packets too early
and unnecessarily bounces the link. Increase this to 50ms taking into
account the switch packet discard timer default and the worst case
per-VL package drainage rate.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The credit return threshold adjustment on mtu change algorithm does not
take into account all the kernel send contexts that are assigned per VL.
Use the pio send context map to adjust the credit return thresholds for
all the allocated and assigned kernel send contexts based on the MTU
adjustment per VL.
The pio send context map can be changed dynamically based on the actual
number of operational vls which is set by the fabric manager. When this
happens update the credit return threshold values for all the remapped
kernel send contexts.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Avoid that sparse reports the following warnings for the hfi1 driver:
trace.c:217:13: warning: no previous prototype for ‘print_u64_array’ [-Wmissing-prototypes]
user_sdma.c:1361:17: warning: dubious: !x & y
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The first argument of test_bit() and clear_bit() is a bit number and
not a bitmask. Hence change that first argument from (1 << 0) into 0.
This patch avoids that smatch reports the following warnings:
user_sdma.c:1059: sdma_cache_evict() warn: test_bit() takes a bit number
user_sdma.c:1590: sdma_rb_remove() warn: test_bit() takes a bit number
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Make the indentation of the source code consistent. Detected by
smatch.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When CONFIG_FRAME_WARN is set to 1024 bytes, which is useful to find
stack consumers, we get a warning in hfi1 driver.
drivers/infiniband/hw/hfi1/affinity.c: In function
‘hfi1_get_proc_affinity’:
drivers/infiniband/hw/hfi1/affinity.c:415:1: warning: the frame size of
1056 bytes is larger than 1024 bytes [-Wframe-larger-than=]
This change removes unneeded buf[1024] declaration and usage.
Fixes: f48ad614c1 ("IB/hfi1: Move driver out of staging")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
That extra tabs are misleading.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The pio map initialization function is off by 1 causing the last
kernel send context that is allocated to not get mapped into the
pio map which leads to the last kernel send context not being used
by any of the qps.
The send context reserved for VL15 is taken care of by setting the
scontext variable that is used as the index into the kernel send
context array to 1 and does not need to be accounted for in the
kernel send context counting loop as it is currently done.
Fix the kernel send context counting loop to account for all the
allocated send contexts and map all of them to the different VLs.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Two 8051 link settings, external device config and tuning method,
were written in the wrong location and the previous settings were
not cleared. For both, clear the old value and write the new
value.
Fixes: 8ebd4cf185 ("staging/rdma/hfi1: Add active and optical cable support")
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When FM is disabled, and the HFI port on the switch is
changed from MgmtAllowed=YES to MgmtAllowed=NO and the
link is bounced, FULL_MGMT_P_KEY doesn't get cleared
from the pkey table. This also occurs when the QSFP
cable is moved from a switch port with MgmtAllowed=YES
to a MgmtAllowed=NO port. Clear pkey entry properly.
Also, when the driver is loaded and the switch port is
set to MgmtAllowed=NO, FULL_MGMT_P_KEY shouldn't be added
to pkey table after FM is started. Only set FULL_MGMT_P_KEY
in the pkey table if switch port is configured to
MgmtAllowed=YES.
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Commit b9b06cb6fe
("IB/hfi1: Fix missing lock/unlock in verbs drain callback")
added a spin lock.
Unfortunately, the new lock code can be called from a base
level interrupt state, and an interrupt that can get stacked
will attempt to get the same lock.
Fix by using the flag save/restore spin lock variation.
Cc: stable@vger.kernel.org # 4.6+
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@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>
Enable trace generation for packets with the "Send Last with
Invalidate" and "Send Only with Invalidate" opcodes.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A new union member "ieth" (Invalidate Extended Transport Header) is
added to the packet header definition in preparation of supporting
the send with invalidate opcode.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The TODO list for the hfi1 driver was completed during 4.6. In addition
other objections raised (which are far beyond what was in the TODO list)
have been addressed as well. It is now time to remove the driver from
staging and into the drivers/infiniband sub-tree.
Reviewed-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
