From: Mitko Haralanov
Expected receives work by user-space libraries (PSM) calling into the
driver with information about the user's receive buffer and have the driver
DMA-map that buffer and program the HFI to receive data directly into it.
This is an expensive
From: Jubin John
The name used to create the verbs txreq cache was not qualified with the unit
number. This causes a panic when destroying the cache on a dual HFI systems.
Create a unique name with the unit number with this patch
Reviewed-by: Mike Marciniszyn
From: Dean Luick
When under heavy load, the send handler can run too long without allowing other
tasks to run. Add a conditional resched to break this up.
Reviewed-by: Mike Marciniszyn
Signed-off-by: Dean Luick
From: Easwar Hariharan
The DC firmware overwrites the enable_lane_tx register and does not update it
on a host request to go to Poll. This causes an infinite loop through the LNI
state machine if a link width downgrade occurs. This patch re-sets the
enable_lane_tx
From: Mike Marciniszyn
Add additional rc traces to aid in debugging rc retry logic.
Reviewed-by: Dennis Dalessandro
Signed-off-by: Mike Marciniszyn
Signed-off-by: Ira Weiny
---
From: Dean Luick
In preparation for threading the receive interrupt, add irqsaves in the packet
processing path.
When the receive interrupt is threaded, the packet processing path is no longer
guaranteed to have IRQs disabled. Add irqsaves where needed on several locks
in
From: Mike Marciniszyn
The workqueue is currently single threaded per port which for a small number of
SDMA engines is ok.
For hfi1, the there are up to 16 SDMA engines that can be fed descriptors in
parallel.
This patch:
- Converts to use alloc_workqueue
- Changes
From: Mitko Haralanov
User SDMA keeps track of progress into the submitted IO vectors by tracking an
offset into the vectors when packets are submitted. This offset is updated
after a successful submission of a txreq to the SDMA engine.
The same offset was used when
From: Ignacio Hernandez
A value of 2048 increased PSM performance while not impacting verbs
performance.
Reviewed-by: Mike Marciniszyn
Reviewed-by: Mitko Haralanov
Signed-off-by: Ignacio Hernandez
From: Jubin John
Signed-off-by: Jubin John
Signed-off-by: Ira Weiny
---
drivers/staging/rdma/hfi1/common.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/staging/rdma/hfi1/common.h
From: Caz Yokoyama
Reset firmware instead of reloading Sbus firmware if it's already done for this
ASIC. To work around thermal polling problem in firmware, don't reload Sbus
firmware, instead, reset the firmware on the initialization of the second HFI.
Reviewed-by:
From: Vennila Megavannan
If packets stop egressing the hardware link, software can lock up.
Implement a timeout for send context halt recovery. This patch increases the
timeout for packet egress to 500 us and timer resets to zero if the packet
occupancy changes.
From: Dean Luick
The latest version of the 8051 firmware will wait longer
when bringing the link down. Extend the driver's timeout
to go with that.
Reviewed-by: Dennis Dalessandro
Signed-off-by: Dean Luick
From: Mitko Haralanov
Clean up the context and sdma macros and move them to a more logical place in
hfi.h
Signed-off-by: Mitko Haralanov
Signed-off-by: Ira Weiny
---
drivers/staging/rdma/hfi1/hfi.h | 22
From: Mike Marciniszyn
The clear_ahg call is new in hfi1 vs qib. For small messages the progress
routine always builds one and clears out the ahg state when the queue has gone
to empty which is the predominant case for small messages.
Inline the routine and avoid
From: Easwar Hariharan
The QSFP interface code has been running without issues and the flag is
never set to off. This patch removes the QSFP_ENABLED bit from HFI1_CAP.
Reviewed-by: Mike Marciniszyn
Signed-off-by: Easwar Hariharan
From: Dean Luick
When under heavy load, the receive interrupt handler can run too long with IRQs
disabled. Add a mixed-mode threading scheme. Initially process packets in the
handler for quick responses (latency). If there are too many packets to
process move to a thread
From: Mitko Haralanov
The address being mapped into a process's memory for notification events was
improperly calculated due to cast that was happening too early. dd->events is a
pointer and should have been casted to unsigned long after the pointer
arithmetic was
From: Niranjana Vishwanathapura
This fixes transmit errors when the number of scatter gather elements in the
request is more that the number of per packet descriptors supported by the
hardware, allocate and coalesce the extra scatter gather elements into a
From: Mitko Haralanov
The SDMA engines were configured to generate progress interrupts every time they
processed N/2 descriptors (where N is the size of the descriptor queue). This
interval was too infrequent, leading to degraded performance.
This commit adds a module
From: Jareer Abdel-Qader
Driver does not verify userid for shared context assignments, allowing
malicious user access.
Reviewed by: Mike Marciniszyn
Signed-off-by: Jareer H Abdel-Qader
Signed-off-by:
From: Ira Weiny
The following are fixes to the hfi1 driver for stability, security, and
performance.
Changes from v2:
Rebased to latest staging-next
Update commit messages
Squashed these patches together:
From: Easwar Hariharan
The ASIC registers were not reset on FLR, and the code to
protect the ASIC block against multiple initializations by
peer HFIs did not extend to multiple ASICs in a system. This
patch addresses this gap.
Reviewed-by: Dean Luick
Hi,
OFED-3.18-1-rc2 is available at:
https://openfabrics.org/downloads/OFED/ofed-3.18-1/OFED-3.18-1-rc2.tgz
To get BUILD_ID run ofed_info
Please report any issues in bugzilla
http://bugs.openfabrics.org/bugzilla/ for OFED 3.18-1
Release notes:
> > Sean, I need to close on this patch. What is your position after
> > Matan's explanation?
> >
>
> Absent an objection from Sean, I've pulled this in. A use after free
> bug is a pretty serious issue, and you've listed an error flow that
> triggers it. The only thing bugging me is that this
25 matches
Mail list logo