- 04 May, 2017 16 commits
-
-
Michael J. Ruhl authored
The error path for context initialization is not consistent. Cleanup all resources on failure. Removed unused variable user_event_mask. Add the _BASE_FAILED bit to the event flags so that a base context can notify waiting sub contexts that they cannot continue. Running out of sub contexts is an EBUSY result, not EINVAL. 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>
-
Michael J. Ruhl authored
The current algorithm for generating sub-context IDs is FILO. If the contexts are not closed in that order, the uniqueness of the ID will be compromised. I.e. logging the creation/deletion of context IDs with an application that assigns and closes in a FIFO order reveals: cache_id: assign: uctxt: 3 sub_ctxt: 0 cache_id: assign: uctxt: 3 sub_ctxt: 1 cache_id: assign: uctxt: 3 sub_ctxt: 2 cache_id: close: uctxt: 3 sub_ctxt: 0 cache_id: assign: uctxt: 3 sub_ctxt: 2 <<< The sub_ctxt ID 2 is reused incorrectly. Update the sub-context ID assign algorithm to use a bitmask of in_use contexts. The new algorithm will allow the contexts to be closed in any order, and will only re-use unused contexts. Size subctxt and subctxt_cnt to match the user API size. Reviewed-by:
-
Michael J. Ruhl authored
Context initialization mixes base context init with sub context init. This is bad because contexts can be reused, and on reuse, reinit things that should not re-initialized. Normalize comments and function names to refer to base context and sub context (not main, shared or slaves). Separate the base context initialization from sub context initialization. hfi1_init_ctxt() cannot return an error so changed to a void and remove error message. Reviewed-by:
-
Michael J. Ruhl authored
In the close path the context is removed from the device array, and then the clear pkey function is called. The pkey function trys to get the context from the device array, but because it was removed the clearing does not occur. Rework pkey clear function to work as expected. Update the function variable to reflect the correct size and name of the hw_context. Reviewed-by:
-
Michael J. Ruhl authored
The search for available shared contexts walks each registered hfi1 device. This search is too broad because other devices may not be on the same fabric, and using its contexts could cause unexpected behavior. Removed walking the list of devices, limiting the search to the opened device. With the device walk removed, the hfi1_devdata (dd) is not available. Added it to the hfi1_filedata for reference. With this change, hfi1_count_units() was rendered obsolete and was removed. Reviewed-by:
-
Sebastian Sanchez authored
The AHG index is only accessed in the request call from user space, so there's no need for atomic semantics. Replace atomic operations for SDMA_REQ_HAVE_AHG bit with a test of the AHG index. Reviewed-by:
Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by:
-
Michael J. Ruhl authored
Since almost all functions that use the hfi1_filedata get the pointer from the file pointer, simplify by only passing the hfi1_filedata pointer. Reviewed-by:
Ira Weiny <ira.weiny@intel.com> Signed-off-by:
-
Michael J. Ruhl authored
To improve the readability of function prototypes, give the parameters names. Reviewed-by:
-
Michael J. Ruhl authored
The only context that frees user_exp_rcv data structures is the last context closed (from a sub-context set). This leaks the allocations from the other sub-contexts. Separate the common frees from the specific frees and call them at the appropriate time. Using KEDR to check for memory leaks we get: Before test: [leak_check] Possible leaks: 25 After test: [leak_check] Possible leaks: 31 (6 leaked data structures) After patch applied (before and after test have the same value) [leak_check] Possible leaks: 25 Each leak is 192 + 13440 + 6720 = 20352 bytes per sub-context. Cc: stable@vger.kernel.org Reviewed-by:
-
Michael J. Ruhl authored
If the eager buffer allocation fails, it is necessary to return an error code. Cc: stable@vger.kernel.org Reviewed-by:
-
Tymoteusz Kielan authored
Performance analysis shows benefits for PSM2 in increasing eager buffer size from 2MB to 8MB. The change has neutral impact on verbs. Make change to the module parameter's default value. Allocation ring down was verified to work with the larger buffer size. Reviewed-by:
Tadeusz Struk <tadeusz.struk@intel.com> Reviewed-by:
Tymoteusz Kielan <tymoteusz.kielan@intel.com> Signed-off-by:
-
Sebastian Sanchez authored
Div instructions show costly in profiles when the tx request header is set. Using right shift instead of a divide operation reduces the cycles spent in the function that sets the tx request header as shown in the profile. Use right shift operation instead. Profile before change: 43.24% 009 | |--23.41%-- user_sdma_send_pkts | | | |--99.90%-- hfi1_user_sdma_process_requestAfter: Profile after change: 45.75% 009 | |--14.81%-- user_sdma_send_pkts | | | |--99.95%-- hfi1_user_sdma_process_request Reviewed-by:
-
Mike Marciniszyn authored
When there are many RC QPs and an RDMA READ request is sent, timeouts occur on the requester side because of fairness among RC QPs on their relative SDMA engine on the responder side. This also hits write and send, but to a lesser extent. Complicating the issue is that the current code checks if workqueue is congested before scheduling other QPs, however, this check is based on the number of active entries in the workqueue, which was found to be too big to for workqueue_congested() to be effective. Fix by reducing the number of active entries as revealed by experimentation from the default of num_sdma to HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES. Retry counts were monitored to determine the correct value. Tracing to investigate any future issues is also added. Reviewed-by:
-
Mike Marciniszyn authored
This field is causing excessive cache line bouncing. There are spare bytes in the r_lock cache line so the best approach is to make an rvt QP field and remove from the hfi1 priv field. Signed-off-by:
-
Jakub Byczkowski authored
In goto_offline() function pstate is masked by 0xff when compared to PLS_OFFLINE state. Mask should be 0xf0, since upper 4 bits specify the "major" state. Reviewed-by:
Easwar Hariharan <easwar.hariharan@intel.com> Signed-off-by:
Jakub Byczkowski <jakub.byczkowski@intel.com> Signed-off-by:
-
Zhu Yanjun authored
In order to let the bonding driver report the correct speed of the underlaying interfaces, when they are IPoIB, the ethtool function get_link_ksettings() in the IPoIB driver is implemented. Cc: Joe Jin <joe.jin@oracle.com> Cc: Junxiao Bi <junxiao.bi@oracle.com> Suggested-by:
Håkon Bugge <Haakon.Bugge@oracle.com> Signed-off-by:
Zhu Yanjun <yanjun.zhu@oracle.com> Signed-off-by:
-
- 02 May, 2017 1 commit
-
-
Paolo Abeni authored
With commit eea40b8f ("infiniband: call ipv6 route lookup via the stub interface"), if the route lookup fails due to ipv6 being disabled, the dst variable is left untouched, and the following dst_release() may access uninitialized memory. Since ipv6_dst_lookup() always sets dst to NULL in case of lookup failure with ipv6 enabled, fix the above just returning the error code if the lookup fails. Fixes: eea40b8f ("infiniband: call ipv6 route lookup via the stub interface") Reported-by:
Sabrina Dubroca <sd@queasysnail.net> Signed-off-by:
Paolo Abeni <pabeni@redhat.com> Reviewed-by:
-
- 01 May, 2017 23 commits
-
-
Doug Ledford authored
When importing the patch 57520751 (IB/SA: Add OPA path record type), a new header file should have been added to the repo as part of the patch. However, as the patch didn't apply cleanly using git am, I instead used patch manually, and followed that up with git add -u, which misses new files. This adds the new file back in. Fixes: 57520751 (IB/SA: Add OPA path record type) Signed-off-by:
-
Tim Wright authored
Add port_xmit_wait to the error counters read by mlx5_ib_process_mad to ensure sysfs port counter provides correct value for PortXmitWait. Otherwise the sysfs port_xmit_wait file always contains zero. The previous MAD_IFC implementation populated this counter, but it was removed during the migration to PPCNT for error counters (32-bit only). Signed-off-by:
Tim Wright <tim@binbash.co.uk> Signed-off-by:
-
Michael Mera authored
In write to debugfs file 'resource_stats' the local buffer 'tmp_str' is written at index 'count-1' where 'count' is the size of the write, so potentially 0. This patch filters odd values for the write size/position to avoid this type of problem. Signed-off-by:
Michael Mera <dev@michaelmera.com> Reviewed-by:
Leon Romanovsky <leonro@mellanox.com> Signed-off-by:
-
Håkon Bugge authored
The last two actual parameters when calling id_map_find_by_sl_id() from id_map_get() are swapped. However, the same formal parameters to id_map_get() have them swapped as well, inverting the effect of the first error. This commit improves readability, but makes no functional change to the code. Signed-off-by:
Wengang Wang <wen.gang.wang@oracle.com> Reviewed-by:
Knut Omang <knut.omang@oracle.com> Acked-by:
-
Håkon Bugge authored
Change and simplify the code to match the variable name. This commit improves readability but makes no functional change to the code. Signed-off-by:
-
Sagi Grimberg authored
In case we got an initial sg_offset, we need to account for it in the mr length. Cc: stable@vger.kernel.org Fixes: ff2ba993 ("IB/core: Add passing an offset into the SG to ib_map_mr_sg") Signed-off-by:
Sagi Grimberg <sagi@grimberg.me> Tested-by:
Israel Rukshin <israelr@mellanox.com> Signed-off-by:
-
Johannes Thumshirn authored
When reading a RDMA WRITE FIRST packet we copy the DMA length from the RDMA header into the qp->resp.resid variable for later use. Later in check_rkey() we clamp it to the MTU if the packet is an RDMA WRITE packet and has a residual length bigger than the MTU. Later in write_data_in() we subtract the payload of the packet from the residual length. If the packet happens to have a payload of exactly the MTU size we end up with a residual length of 0 despite the packet not being the last in the conversation. When the next packet in the conversation arrives, we don't have any residual length left and thus set the QP into an error state. This broke NVMe over Fabrics functionality over rdma_rxe.ko The patch was verified using the following test. # echo eth0 > /sys/module/rdma_rxe/parameters/add # nvme connect -t rdma -a 192.168.155.101 -s 1023 -n nvmf-test # mkfs.xfs -fK /dev/nvme0n1 meta-data=/dev/nvme0n1 isize=256 agcount=4, agsize=65536 blks = sectsz=4096 attr=2, projid32bit=1 = crc=0 finobt=0, sparse=0 data = bsize=4096 blocks=262144, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0 ftype=1 log =internal log bsize=4096 blocks=2560, version=2 = sectsz=4096 sunit=1 blks, lazy-count=1 realtime =none extsz=4096 blocks=0, rtextents=0 # mount /dev/nvme0n1 /tmp/ [ 148.923263] XFS (nvme0n1): Mounting V4 Filesystem [ 148.961196] XFS (nvme0n1): Ending clean mount # dd if=/dev/urandom of=test.bin bs=1M count=128 128+0 records in 128+0 records out 134217728 bytes (134 MB, 128 MiB) copied, 0.437991 s, 306 MB/s # sha256sum test.bin cde42941f045efa8c4f0f157ab6f29741753cdd8d1cff93a6b03649d83c4129a test.bin # cp test.bin /tmp/ sha256sum /tmp/test.bin cde42941f045efa8c4f0f157ab6f29741753cdd8d1cff93a6b03649d83c4129a /tmp/test.bin Signed-off-by:Johannes Thumshirn <jthumshirn@suse.de> Cc: Hannes Reinecke <hare@suse.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Max Gurtovoy <maxg@mellanox.com> Acked-by:
Moni Shoua <monis@mellanox.com> Signed-off-by:
-
Dasaratharaman Chandramouli authored
When the bit 26 of capmask2 field in OPA classport info query is set, SA will query for OPA path records instead of querying for IB path records. Note that OPA path records can only be queried by kernel ULPs. Userspace clients continue to query IB path records. Reviewed-by:
Don Hiatt <don.hiatt@intel.com> Reviewed-by:
Dasaratharaman Chandramouli <dasaratharaman.chandramouli@intel.com> Signed-off-by:
-
Dasaratharaman Chandramouli authored
Add opa_sa_path_rec to sa_path_rec data structure. The 'type' field in sa_path_rec identifies the type of the path record. Reviewed-by:
-
Dasaratharaman Chandramouli authored
sa_path_rec now contains a union of sa_path_rec_ib and sa_path_rec_roce based on the type of the path record. Note that fields applicable to path record type ROCE v1 and ROCE v2 fall under sa_path_rec_roce. Accessor functions are added to these fields so the caller doesn't have to know the type. Reviewed-by:
-
Dasaratharaman Chandramouli authored
struct sa_path_rec has a gid_type field. This patch introduces a more generic path record specific type 'rec_type' which is either IB, ROCE v1 or ROCE v2. The patch also provides conversion functions to get a gid type from a path record type and vice versa Reviewed-by:
-
Dasaratharaman Chandramouli authored
Rename ib_sa_path_rec to a more generic sa_path_rec. This is part of extending ib_sa to also support OPA path records in addition to the IB defined path records. Reviewed-by:
-
Dasaratharaman Chandramouli authored
This patch adds braces around parameters to sizeof as called out by checkpatch Reviewed-by:
-
Dasaratharaman Chandramouli authored
OPA ah_attr types allows core components to specify attributes that may be specific to opa devices. For instance, opa type ah_attr provides 32 bit lids enabling larger OPA fabric sizes. Reviewed-by:
Sean Hefty <sean.hefty@intel.com> Signed-off-by:
-
Dasaratharaman Chandramouli authored
rdma_ah_attr can now be either ib or roce allowing core components to use one type or the other and also to define attributes unique to a specific type. struct ib_ah is also initialized with the type when its first created. This ensures that calls such as modify_ah dont modify the type of the address handle attribute. Reviewed-by:
Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com> Signed-off-by:
-
Dasaratharaman Chandramouli authored
Modify core and driver components to use accessor functions introduced to access individual fields of rdma_ah_attr Reviewed-by:
-
Dasaratharaman Chandramouli authored
These accessor functions are supposed to be used to get and set individual fields of struct rdma_ah_attr Reviewed-by:
-
Dasaratharaman Chandramouli authored
Functions pvrdma_ah_attr_to_ib and ib_ah_attr_to_pvrdma have been renamed so they are in sync wit the rename of the ib_ah_attr structure Reviewed-by:
-
Dasaratharaman Chandramouli authored
local function to_ib_ah_attr is renamed so it in sync with the rename of the ib_ah_attr structure Reviewed-by:
-
Dasaratharaman Chandramouli authored
local function to_ib_ah_attr is renamed so it in sync with the rename of the ib_ah_attr structure Reviewed-by:
-
Dasaratharaman Chandramouli authored
local function to_ib_ah_attr is renamed so it in sync with the rename of the ib_ah_attr structure Reviewed-by:
-
Dasaratharaman Chandramouli authored
Rename ib_destroy_ah to rdma_destroy_ah so its in sync with the rename of the ib address handle attribute Reviewed-by:
-
Dasaratharaman Chandramouli authored
Rename ib_query_ah to rdma_query_ah so its in sync with the rename of the ib address handle attribute Reviewed-by:
-
