Mike Marciniszyn [Thu, 12 May 2016 17:22:57 +0000 (10:22 -0700)]
IB/hfi1: Fix pio wait counter double increment
The code unconditionlly increments the pio wait counter
making the counter inacurate and unusable.
Fixes: 14553ca11039 ("staging/rdma/hfi1: Adaptive PIO for short messages") Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
The external device configuration was incorrectly shifted to byte 3 of
the 32 bit DC_HOST_COMM_SETTINGS instead of byte 0. This patch corrects
the shift and provides the cable capability information in byte 0.
Easwar Hariharan [Thu, 12 May 2016 17:22:39 +0000 (10:22 -0700)]
IB/hfi1: Wait for QSFP modules to initialize
The function level reset in init_chip() and subsequent write of all 1s
to the ASIC_QSFP registers effectively resets attached active and
optical QSFP modules that pay attention to the RESET_N pin.
We subsequently try to access the QSFP management interface to qualify
and tune the channel and fabric SerDes before enough time (2 seconds
per SFF 8679 spec for QSFP28 modules) has elapsed for the module to
finish initialization. This fails and causes the failure of the channel
tuning algorithm, preventing us from bringing the link up.
This patch checks the port type prior to beginning channel and SerDes
tuning, and if found to be QSFP, watches for the QSFP initialization
complete interrupt, with a maximum timeout of 2 seconds, to allow the
initialization to complete.
Easwar Hariharan [Thu, 12 May 2016 17:22:33 +0000 (10:22 -0700)]
IB/hfi1: Ignore non-temperature warnings on a downed link
QSFP modules can raise an interrupt to inform us of expected conditions
while the link is down, such as RX power low. Actively ignore these
conditions when the link is down as they only add reporting noise.
Continue reporting conditions that are valid at all times, such as
temperature alarms and warnings.
Adding the needed mlx5_ifc hardware bits and structs
for the following features:
* Add vport to steering commands for SRIOV ACL support
* Add mlcr, pcmr and mcia registers for dump module EEPROM
* Add support for FCS, beacon led and disable_link bits to
hca caps
* Add CQE period mode bit in CQ context for CQE based CQ
moderation support
* Add umr SQ bit for fragmented memory registration
* Add needed bits and caps for Striding RQ support
In-order to avoid possible future conflicts between rdma and
net-next we added all expected updates to this file for this release.
If more changes will be submitted, we plan to do it only through
one of the subsystems, probably net-next.
All updated bits in this patch will be later used in
the up-coming submissions to net-next and rdma trees.
Since all srp_map_finish_fr() callers pass a non-zero value as
the fourth argument (sg_nents), the sg_nents == 0 check in that
function can be removed. Add a count == 0 check in the caller
of that function.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Laurence Oberman <loberman@redhat.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/srp: Avoid that mapping failure triggers an infinite loop
The srp_queuecommand() function translates ENOMEM into QUEUE_FULL
which causes the SCSI mid-layer to retry the command. All other
error codes are translated into DID_ERROR which causes the SCSI
command to fail. Return E2BIG if mapping will always fail to
prevent that the SCSI mid-layer keeps resubmitting a command
forever.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Laurence Oberman <loberman@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
Ensure that req->nmdesc is set correctly in srp_map_sg() if mapping
fails. Avoid that mapping failure causes a memory descriptor leak.
Report srp_map_sg() failure to the caller.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Laurence Oberman <loberman@redhat.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
Hariprasad S [Wed, 4 May 2016 19:57:36 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: move QP -> ERROR on fatal disconnect errors
In c4iw_ep_disconnect(), if we fail to initiate a close operation, then
move the qp to ERROR to disassociate the ep from the qp. Failure to do
this will leak the ep resources.
Hariprasad S [Wed, 4 May 2016 19:57:35 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: don't use abort_connection in process_mpa_request()
Instead return whether the caller needs to disconnect. This is part of
getting rid of abort_connection() altogether so we properly clean up on
send_abort() failures.
Hariprasad S [Wed, 4 May 2016 19:57:32 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: remove connection abort from process_mpa_reply
Instead, have the caller, rx_data() handle the close/abort like
it does for process_mpa_request(). This is part of getting rid of
abort_connection() altogether so we properly clean up on send_abort()
failures.
Hariprasad S [Wed, 4 May 2016 19:57:31 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: ensure eps don't get freed while the mutex is held
In rx_data(), with the ep in FPDU_MODE, refcnt=2, if we get unexpected
streaming data, we call c4iw_modify_rc_qp() and move the qp from
RTS -> TERMINATE. In c4iw_modify_rc_qp(), if rdma_fini() returns
an error, the ep will be dereferenced (refcnt=1). Then rx_data()
calls c4iw_ep_disconnect() which starts the close operation.
But if send_halfclose() fails in c4iw_ep_disconnect(), we will call
release_ep_resources() derefing the ep which reduces the refcnt to 0 and
and frees the ep. However we still has the ep mutex at that point, so we
have a touch-after-free bug. There is a similar issue where
peer_close() calls c4iw_ep_disconnect().
The solution is to add a reference to the ep in c4iw_ep_disconnect()
after acquiring the mutex, and release it after releasing the mutex.
Hariprasad S [Wed, 4 May 2016 19:57:30 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: stop ep timer on close failure
In c4iw_ep_disconnect(), if we start the ep timer to begin a close,
but send_halfclose() fails, we need to stop the timer and send a CLOSE
event up to the IWCM before releasing the resources. Otherwise, we can
crash when the ep timer fires if the ep is referencing a previous instance
of the device. This can happen as part of adapter reset/recovery, for
instance.
Hariprasad S [Wed, 4 May 2016 19:57:29 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: release ep resources on accept arp failure
If ARP fails before the CPL_PASS_ACCEPT_RPL is seen by hardware, the tid
will be stuck in SYN_PEND and never released. So create an arp failure
handler specifically for this message to release the endpoint resources.
In pass_accept_rpl_arp_failure(), put the parent endpoint so it will
be freed when destroyed. Also we don't need to call release_tid() here
because _c4iw_free_ep() calls cxgb4_remove_tid() which releases the
hwtid.
If we get an ABORT_REQ_RSS instead of a PASS_ESTABLISH (because the
peer's ACK to our SYN is never received), then put the parent as well
in peer_abort().
Treat accept_cr() failures just like arp failures: put the parent ep
and release the ep resources destroying the tid
The ARP failure handlers are called in an atomic context, so we need to
schedule some of the processing which might block. Namely _c4iw_free_ep()
which needs a mutex. So create a "special" CPL opcode and handler and
schedule it via sched() to be run by process_work() in a blockable context.
Also rework the active open arp failure handler to make use of
release_ep_resources(). This allows both the active and passive arp
failure handlers to use the same deferred cleanup function.
iSER currently has a couple places that set max_sectors in either the host
template or SCSI host, and all of them get it wrong.
This patch instead uses a single assignment that (hopefully) gets it right:
the max_sectors value must be derived from the number of segments in the
FR or FMR structure, but actually be one lower than the page size multiplied
by the number of sectors, as it has to handle the case of non-aligned I/O.
Without this I get trivial to reproduce hangs when running xfstests
(on XFS) over iSER to Linux targets.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Max Gurtovoy <maxg@mellanox.com> Acked-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Doug Ledford <dledford@redhat.com>
RDMA/i40iw: Fix for checking if the QP is destroyed
Fix for checking if the QP associated with a completion
has been destroyed while processing CQ elements.
If that is the case, move the CQ head to the next element
and continue completion processing.
STag index mask is calculated incorrectly, missing
the 14 bits minimum requirement. Add max macro to use
either # of MRs or 14 bits in the mask size calculation.
Ismail, Mustafa [Mon, 18 Apr 2016 15:33:09 +0000 (10:33 -0500)]
RDMA/i40iw: Adding queue drain functions
Adding sq and rq drain functions, which block until all
previously posted wr-s in the specified queue have completed.
A completion object is signaled to unblock the thread,
when the last cqe for the corresponding queue is processed.
Ismail, Mustafa [Mon, 18 Apr 2016 15:33:08 +0000 (10:33 -0500)]
RDMA/i40iw: Fix SD calculation for initial HMC creation
Correct SD calculation by using base address returned from commit FPM.
This alleviates any assumptions on resource ordering and alignment
requirement. Also consolidate SD estimation code into i40iw_est_sd().
Jubin John [Thu, 14 Apr 2016 15:31:53 +0000 (08:31 -0700)]
IB/hfi1: Serialize hrtimer function calls
hrtimer functions do not guarantee serialization, so we extend the
cca_timer_lock to cover the hrtimer_forward_now() in the hrtimer
callback handler and the hrtimer_start() in process_becn(). This
prevents races between these 2 functions to update the hrtimer state
leading to problems such as:
kernel BUG at kernel/hrtimer.c:1282!
encountered during validation of the CCA feature.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jubin John <jubin.john@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Dean Luick [Thu, 14 Apr 2016 15:31:42 +0000 (08:31 -0700)]
IB/hfi1: Correctly report neighbor link down reason
The code to save the link down reason for reporting to the SMA
was in a location before the actual reason was read. Move the
SMA link down reason assignment to a better location.
Dean Luick [Thu, 14 Apr 2016 15:31:36 +0000 (08:31 -0700)]
IB/hfi1: Use the neighbor link down reason only when valid
The 8051 uses a link down reason to inform the driver why the
link went down. The neighbor planned link down reason code is
only valid when a link down idle message is received by the 8051.
Enhance the explanation on why the link went down.
Dean Luick [Thu, 14 Apr 2016 15:31:30 +0000 (08:31 -0700)]
IB/hfi1: Ignore link downgrade with 0 lanes
Versions of the 8051 firmware < 0.38 may report a link failure
as a link downgrade with a width of 0 followed by a link down
notification. Ignore the zero width downgrade notification -
the driver should follow the link down path.
Dean Luick [Tue, 12 Apr 2016 18:32:06 +0000 (11:32 -0700)]
IB/hfi1: Add RSM rule for user FECN handling
Add a receive side mapping rule to extract expected user packets with
the FECN bit set and place them in an eager buffer. This will allow
user libraries to recognize that a FECN was sent when using header
suppression and respond appropriately.
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>
Dean Luick [Tue, 12 Apr 2016 18:31:11 +0000 (11:31 -0700)]
IB/hfi1: Move QOS decision logic into its own function
The decision to use QOS affects other resource allocation.
Move the QOS decision logic into its own function so it can
be called by other interested parties.
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>
Dean Luick [Tue, 12 Apr 2016 18:30:51 +0000 (11:30 -0700)]
IB/hfi1: Extract RSM map table init from QOS
Refactor the allocation, tracking, and writing of the RSM map table
into its own set of routines. This will allow the map table to be
passed to multiple users to fill in as needed. Start with the original
user, QOS.
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>
IB/hfi1: Reduce kernel context pio buffer allocation
The pio buffers were pooled evenly among all kernel contexts and
user contexts. However, the demand from kernel contexts is much
lower than user contexts. This patch reduces the allocation for
kernel contexts and thus makes more credits available for PSM,
helping performance. This is especially useful on high core-count
systems where large numbers of contexts are used.
A new context type SC_VL15 is added to distinguish the context used
for VL15 from other kernel contexts. The reason is that VL15 needs
to support 2KB sized packet while other kernel contexts need only
support packets up to the size determined by "piothreshold", which
has a default value of 256.
The new allocation method allows triple buffering of largest pio
packets configured for these contexts. This is sufficient to maintain
verbs performance. The largest pio packet size is 2048B for VL15
and "piothreshold" for other kernel contexts. A cap is applied to
"piothreshold" to avoid excessive buffer allocation.
The special case that SDMA is disable is handled differently. In
that case, the original pooling allocation is used to better
support the much higher pio traffic.
Notice that if adaptive pio is disabled (piothreshold==0), the pio
buffer size doesn't matter for non-VL15 kernel send contexts when
SDMA is enabled because pio is not used at all on these contexts
and thus the new allocation is still valid. If SDMA is disabled then
pooling allocation is used as mentioned in previous paragraph.
Adjustment is also made to the calculation of the credit return
threshold for the kernel contexts. Instead of purely based on
the MTU size, a percentage based threshold is also considered and
the smaller one of the two is chosen. This is necessary to ensure
that with the reduced buffer allocation credits are returned in
time to avoid unnecessary stall in the send path.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Reviewed-by: Dean Luick <dean.luick@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Mark Debbage <mark.debbage@intel.com> Reviewed-by: Jubin John <jubin.john@intel.com> Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Jubin John [Tue, 12 Apr 2016 18:30:08 +0000 (11:30 -0700)]
IB/hfi1: Change default number of user contexts
Change the default number of user contexts to the number of real
(non-HT) cpu cores in order to reduce the division of hfi1 hardware
contexts in the case of high core counts with hyper-threading enabled.
Reviewed-by: Dean Luick <dean.luick@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com> Signed-off-by: Jubin John <jubin.john@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Mike Marciniszyn [Tue, 12 Apr 2016 18:28:56 +0000 (11:28 -0700)]
IB/hfi1: Remove unreachable code
Remove unreachable code from RC ack handling to fix an
smatch error.
Fixes: 633d27399514 ("staging/rdma/hfi1: use mod_timer when appropriate") Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Dean Luick [Tue, 12 Apr 2016 18:28:36 +0000 (11:28 -0700)]
IB/hfi1: Fix double QSFP resource acquire on cache refresh
The function refresh_qsfp_cache() acquires the i2c chain resource,
but one caller already holds the resource. Change the acquire so
all calls to refresh_qsfp_cache() are covered by the acquire and
remove the acquire within refresh_qsfp_cache().
Dean Luick [Tue, 12 Apr 2016 18:26:21 +0000 (11:26 -0700)]
IB/hfi1: Guard against concurrent I2C access across all chains
The discrete ASIC board design makes the two I2C chains not
independent of each other. That is, only one chain can safely
be accessed at a time. For discrete ASIC devices, adjust the
resource locking so that access to one I2C chain will lock both
of the chains.
The pre-LNI SerDes and channel tuning algorithm already checks for
module presence assertion for the relevant port types. The extraneous
check removed in this patch blocks link up for port types for which
the module presence assertion is not relevant.
IB/hfi1: Always turn on CDRs for low power QSFP modules
Clock and data recovery mechanisms (CDRs) in active QSFP modules
can be turned on or off to improve the bit error rate observed on
the channel. Signal integrity and bit error rate requirements require
us to always turn on any CDRs present in low power cables (power
dissipation 2.5W or lower). However, we adhere to the platform
designer's settings (provided in the platform configuration) for
higher power cables (dissipation 3.5W or higher) if the platform
designer has determined that the platform requires the CDRs to be
turned on (or off) and is capable of supplying and cooling the higher
power modules.
This patch also introduces the get_qsfp_power_class function to
centralize the bit twiddling required to determine the QSFP power class
across the code. Reusing this function improves the readability of code
that depends on knowing the power class of the cable, such as the
active and optical channel tuning algorithm.
Reviewed-by: Dean Luick <dean.luick@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Easwar Hariharan <easwar.hariharan@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/hfi1: Check P_KEY for all sent packets from user mode
Add the P_KEY check for user-context mechanism for
both PIO and SDMA. For PIO, the
SendCtxtCheckEnable.DisallowKDETHPackets is set by
default. When the P_KEY is set,
SendCtxtCheckEnable.DisallowKDETHPackets is cleared.
For SDMA, a software check was included. This change
requires user processes to set the P_KEY before sending
any packets, otherwise, the sent packet will fail. The
original submission didn't have this check but it's
required.
Reviewed-by: Dean Luick <dean.luick@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Mikto Haralanov <mitko.haralanov@intel.com> Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Increasing the default MTU size to 10KB improves performance
for PSM. Change the default MTU to 10KB but constrain
Verbs MTU to 8KB. Also update default MTU module parameter
description to be HFI1_DEFAULT_MAX_MTU.
Reviewed-by: Dean Luick <dean.luick@intel.com> Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Reviewed-by: Jubin John <jubin.john@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>
Dean Luick [Tue, 12 Apr 2016 17:50:35 +0000 (10:50 -0700)]
IB/hfi1: Simplify init_qpmap_table()
Make init_qpmap_table() easier to understand by simplifying
the loop indexing and writing each register when it is "full",
removing the need for a follow-on register write.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Dean Luick [Tue, 12 Apr 2016 17:50:16 +0000 (10:50 -0700)]
IB/hfi1: Remove invalid QOS check
Remove an invalid compare of the number of QOS RSM map table entries
against the number of physical receive contexts. The RSM map table
has its own size and has no relation to the number of physical receive
contexts.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Dean Luick [Tue, 12 Apr 2016 17:50:10 +0000 (10:50 -0700)]
IB/hfi1: Fix QOS num_vl bit width
The bit width for num_vls, n, needs to be calculated based on
the pow2 rounded up of the number of vls. Otherwise num_vls of 3,
5, 6, and 7 will have misplaced QOS RSM map entries.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Dean Luick [Tue, 12 Apr 2016 17:50:04 +0000 (10:50 -0700)]
IB/hfi1: Fix i2c resource reservation checks
The i2c and qsfp read/write routines should check for the resource
reservation of the incoming argument target rather than the implicit
target of the hardware HFI.
Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com> Signed-off-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Jubin John <jubin.john@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Jubin John [Wed, 20 Apr 2016 13:05:24 +0000 (06:05 -0700)]
IB/rdmavt,hfi1,qib: Fix memory leak
rdi->ports has memory allocated in rvt_alloc_device(), but does not get
freed because the hfi1 and qib drivers drivers call ib_dealloc_device()
directly instead of going through rdmavt. Add a rvt_dealloc_device()
that frees rdi->ports and then calls ib_dealloc_device(). Switch hfi1
and qib drivers to calling rvt_dealloc_device() instead of
ib_dealloc_device() directly.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Brian Welty <brian.welty@intel.com> Signed-off-by: Jubin John <jubin.john@intel.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/hfi1: Fix buffer cache races which may cause corruption
There are two possible causes for node/memory corruption both
of which are related to the cache eviction algorithm. One way
to cause corruption is due to the asynchronous nature of the
MMU invalidation and the locking used when invalidating node.
The MMU invalidation routine would temporarily release the
RB tree lock to avoid a deadlock. However, this would allow
the eviction function to take the lock resulting in the removal
of cache nodes.
If the node being removed by the eviction code is the same as
the node being invalidated, the result is use after free.
The same is true in the other direction due to the temporary
release of the eviction list lock in the eviction loop.
Another corner case exists when dealing with the SDMA buffer
cache that could cause memory corruption of kernel memory.
The most common way, in which this corruption exhibits itself
is a linked list node corruption. In that case, the kernel will
complain that a node with poisoned pointers is being removed.
The fact that the pointers are already poisoned means that the
node has already been removed from the list.
To root cause of this corruption was a mishandling of the
eviction list maintained by the driver. In order for this
to happen four conditions need to be satisfied:
1. A node describing a user buffer already exists in the
interval RB tree,
2. The beginning of the current user buffer matches that
node but is bigger. This will cause the node to be
extended.
3. The amount of cached buffers is close or at the limit
of the buffer cache size.
4. The node has dropped close to the end of the eviction
list. This will cause the node to be considered for
eviction.
If all of the above conditions have been satisfied, it is
possible for the eviction algorithm to evict the current node,
which will free the node without the driver knowing.
To solve both issues described above:
- the locking around the MMU invalidation loop and cache
eviction loop has been improved so locks are not released in
the loop body,
- a new RB function is introduced which will "atomically" find
and remove the matching node from the RB tree, preventing the
MMU invalidation loop from touching it, and
- the node being extended by the pin_vector_pages() function is
removed from the eviction list prior to calling the eviction
function.
IB/hfi1: Extract and reinsert MMU RB node on lookup
The page pinning function, which also maintains the pin cache,
behaves one of two ways when an exact buffer match is not found:
1. If no node is not found (a buffer with the same starting address
is not found in the cache), a new node is created, the buffer
pages are pinned, and the node is inserted into the RB tree, or
2. If a node is found but the buffer in that node is a subset of
the new user buffer, the node is extended with the new buffer
pages.
Both modes of operation require (re-)insertion into the interval RB
tree.
When the node being inserted is a new node, the operations are pretty
simple. However, when the node is already existing and is being
extended, special care must be taken.
First, we want to guard against an asynchronous attempt to
delete the node by the MMU invalidation notifier. The simplest way to
do this is to remove the node from the RB tree, preventing the search
algorithm from finding it.
Second, the node needs to be re-inserted so it lands in the proper place
in the tree and the tree is correctly re-balanced. This also requires
the node to be removed from the RB tree.
This commit adds the hfi1_mmu_rb_extract() function, which will search
for a node in the interval RB tree matching an address and length and
remove it from the RB tree if found. This allows for both of the above
special cases be handled in a single step.
Reviewed-by: Dean Luick <dean.luick@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 computation of the interval of an interval RB node
was incorrect leading to data corruption due to the RB
search algorithm not properly finding the all RB nodes
in an MMU invalidation interval.
The problem stemmed from the fact that the beginning
address of the node's range was being aligned to a page
boundary. For certain buffer sizes, this would lead to
a end address calculation that was off by 1 page.
An important aspect of keeping the RB same is also
updating the node's range in the case it's being extended.
IB/hfi1: Protect the interval RB tree when cleaning up
The current implementation of the clean up function for
the interval RB trees has two flaws which may cause
problems in cases of concurrent executing of the function
and MMU notifier.
The flaws were due to the fact that deregistration of the
MMU callbacks was done after the tree was emptied and,
furthermore, the tree was not being locked.
This commit fixes both of these flaws by, first, switch the
order of operations, and, second, locking the tree while
traversing it to prevent any other operations.
The driver had two memory leaks - one in the user
expected receive code and one in SDMA buffer cache.
The leak in the expected receive code only showed up
when the user/admin had set ulimit sufficiently low
and the driver did not have enough room in the cache
before hitting the limit of allowed cachable memory.
When this condition occurred, the driver returned
early signaling userland that it needed to free some
buffers to free up room in the cache.
The bug was that the driver was not cleaning up
allocated memory prior to returning early.
The leak in the SDMA buffer cache could occur (even
though it never did), when the insertion of a buffer
node in the interval RB tree failed. In this case, the
driver failed to unpin the pages of the node instead
erroneously returning success.
IB/hfi1: Don't remove list entries if they are not in a list
The SDMA cache logic maintains an eviction list which is ordered
by most recently used user buffers. Upon errors or buffer freeing,
the list nodes were unconditionally being deleted. This would lead
to list corruption warnings if the nodes were never inserted in the
eviction list to begin with.
This commit prevents this by checking that the nodes are already
part of the eviction list.
Reviewed-by: Dean Luick <dean.luick@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>
Mike Marciniszyn [Tue, 12 Apr 2016 17:46:10 +0000 (10:46 -0700)]
IB/qib, IB/hfi1: Fix up UD loopback use of irq flags
The dual lock patch moved locking around and missed an issue
with handling irq flags when processing UD loopback
packets. This issue was revealed by smatch.
Fix for both qib and hfi1 to pass the saved flags to the UD request
builder and handle the changes correctly.
Fixes: 46a80d62e6e0 ("IB/qib, staging/rdma/hfi1: add s_hlock for use in post send") Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Jason Gunthorpe [Mon, 11 Apr 2016 01:13:13 +0000 (19:13 -0600)]
IB/security: Restrict use of the write() interface
The drivers/infiniband stack uses write() as a replacement for
bi-directional ioctl(). This is not safe. There are ways to
trigger write calls that result in the return structure that
is normally written to user space being shunted off to user
specified kernel memory instead.
For the immediate repair, detect and deny suspicious accesses to
the write API.
For long term, update the user space libraries and the kernel API
to something that doesn't present the same security vulnerabilities
(likely a structured ioctl() interface).
The impacted uAPI interfaces are generally only available if
hardware from drivers/infiniband is installed in the system.
Reported-by: Jann Horn <jann@thejh.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
[ Expanded check to all known write() entry points ] Cc: stable@vger.kernel.org Signed-off-by: Doug Ledford <dledford@redhat.com>
Dean Luick [Fri, 22 Apr 2016 18:17:03 +0000 (11:17 -0700)]
IB/hfi1: Use kernel default llseek for ui device
The ui device llseek had a mistake with SEEK_END and did
not fully follow seek semantics. Correct all this by
using a kernel supplied function for fixed size devices.
Cc: Al Viro <viro@ZenIV.linux.org.uk> 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>
This commit re-arranges the context initialization code in a way that
would allow for context event flags to be used to determine whether
the context has been successfully initialized.
In turn, this can be used to skip the resource de-allocation if they
were never allocated in the first place.
Fixes: 3abb33ac6521 ("staging/hfi1: Add TID cache receive init and free funcs") Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com. Signed-off-by: Doug Ledford <dledford@redhat.com>
Jubin John [Tue, 12 Apr 2016 17:47:00 +0000 (10:47 -0700)]
IB/rdmavt: Fix send scheduling
call_send is used to determine whether to send immediately or schedule
a send for later. The current logic in rdmavt is inverted and has a
negative impact on the latency of the hfi1 and qib drivers. Fix this
regression by correctly calling send immediately when call_send is set.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jubin John <jubin.john@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/hfi1: Fix deadlock caused by locking with wrong scope
The locking around the interval RB tree is designed to prevent
access to the tree while it's being modified. The locking in its
current form is too overzealous, which is causing a deadlock in
certain cases with the following backtrace:
As evident from the backtrace above, the process was being put to sleep
while holding the lock.
Limiting the scope of the lock only to the RB tree operation fixes the
above error allowing for proper locking and the process being put to
sleep when needed.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/hfi1: Prevent NULL pointer deferences in caching code
There is a potential kernel crash when the MMU notifier calls the
invalidation routines in the hfi1 pinned page caching code for sdma.
The invalidation routine could call the remove callback
for the node, which in turn ends up dereferencing the
current task_struct to get a pointer to the mm_struct.
However, the mm_struct pointer could be NULL resulting in
the following backtrace:
To correct this, the mm_struct passed to us by the MMU notifier is
used (which is what should have been done to begin with). This avoids
the broken derefences and ensures that the correct mm_struct is used.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Reviewed-by: Dean Luick <dean.luick@intel.com> Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Sagi Grimberg [Thu, 31 Mar 2016 16:03:25 +0000 (19:03 +0300)]
IB/mlx5: Expose correct max_sge_rd limit
mlx5 devices (Connect-IB, ConnectX-4, ConnectX-4-LX) has a limitation
where rdma read work queue entries cannot exceed 512 bytes.
A rdma_read wqe needs to fit in 512 bytes:
- wqe control segment (16 bytes)
- rdma segment (16 bytes)
- scatter elements (16 bytes each)
So max_sge_rd should be: (512 - 16 - 16) / 16 = 30.
Cc: linux-stable@vger.kernel.org Reported-by: Christoph Hellwig <hch@lst.de> Tested-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Sagi Grimberg <sagig@grimberg.me> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Merge branch 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
Pull workqueue fix from Tejun Heo:
"So, it turns out we had a silly bug in the most fundamental part of
workqueue for a very long time. AFAICS, this dates back to pre-git
era and has quite likely been there from the time workqueue was first
introduced.
A work item uses its PENDING bit to synchronize multiple queuers.
Anyone who wins the PENDING bit owns the pending state of the work
item. Whether a queuer wins or loses the race, one thing should be
guaranteed - there will soon be at least one execution of the work
item - where "after" means that the execution instance would be able
to see all the changes that the queuer has made prior to the queueing
attempt.
Unfortunately, we were missing a smp_mb() after clearing PENDING for
execution, so nothing guaranteed visibility of the changes that a
queueing loser has made, which manifested as a reproducible blk-mq
stall.
Lots of kudos to Roman for debugging the problem. The patch for
-stable is the minimal one. For v3.7, Peter is working on a patch to
make the code path slightly more efficient and less fragile"
* 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
workqueue: fix ghost PENDING flag while doing MQ IO
Merge branch 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup
Pull cgroup fixes from Tejun Heo:
"Two patches to fix a deadlock which can be easily triggered if memcg
charge moving is used.
This bug was introduced while converting threadgroup locking to a
global percpu_rwsem and is caused by cgroup controller task migration
path depending on the ability to create new kthreads. cpuset had a
similar issue which was fixed by performing heavy-lifting operations
asynchronous to task migration. The two patches fix the same issue in
memcg in a similar way. The first patch makes the mechanism generic
and the second relocates memcg charge moving outside the migration
path.
Given that we don't want to perform heavy operations while
writelocking threadgroup lock anyway, moving them out of the way is a
desirable solution. One thing to note is that the problem was
difficult to debug because lockdep couldn't figure out the deadlock
condition. Looking into how to improve that"
* 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
memcg: relocate charge moving from ->attach to ->post_attach
cgroup, cpuset: replace cpuset_post_attach_flush() with cgroup_subsys->post_attach callback
projects / linux-beck.git / log