- 04 Mar, 2016 10 commits
-
-
Chaehyun Lim authored
This patch adds a new enum cfg_type_cmd to change hard-coded command type. Signed-off-by:
Chaehyun Lim <chaehyun.lim@gmail.com> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
Chaehyun Lim authored
This patch renames hardwareProductVersion to hw_product_version to avoid camelcase. Signed-off-by:
-
Chaehyun Lim authored
This patch removes commented codes in struct wilc_cfg_str. Signed-off-by:
-
Chaehyun Lim authored
TAG_PARAM_OFFSET is defined at top of this file so that it is used to simplify codes. Signed-off-by:
-
Anchal Jain authored
Remove a whitespace before parenthesis "(" Signed-off-by:Anchal Jain <anchalj109@gmail.com> Signed-off-by:
-
Chaehyun Lim authored
This patch removes typedef from enum tenuConnectSts and renames it to connect_status to avoid camelcase. Signed-off-by:
-
Chaehyun Lim authored
This patch removes INFINITE_SLEEP_TIME that is not used in the driver, so just remove it. Signed-off-by:
-
Chaehyun Lim authored
This patch removes typedef from struct sdio_cmd53_t and renames it to sdio_cmd53. Signed-off-by:
-
Chaehyun Lim authored
This patch removes unnecessary comment code in struct sdio_cmd53_t. Signed-off-by:
-
Chaehyun Lim authored
This patch removes typedef from struct sdio_cmd52_t and renames it to sdio_cmd52. Signed-off-by:
-
- 03 Mar, 2016 7 commits
-
-
Liang Zhen authored
ko2iblnd may retry too frequent for growing pools, all schedulers are spinning if another thread is in progress of allocating a new pool and can't finish right away because of high system load. Signed-off-by:
Liang Zhen <liang.zhen@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7054 Reviewed-on: http://review.whamcloud.com/16470Reviewed-by:
Doug Oucharek <doug.s.oucharek@intel.com> Reviewed-by:
James Simmons <uja.ornl@yahoo.com> Reviewed-by:
Oleg Drokin <oleg.drokin@intel.com> Signed-off-by:
-
Liang Zhen authored
When there is a connection race between two nodes and one side of the connection is rejected by the other side. o2iblnd will reconnect immediately, this is going to generate a lot of trashes if: - race winner is slow and can't send out connecting request in short time. - remote side leaves a cmid in TIMEWAIT state, which will reject future connection requests To resolve this problem, this patch changed the reconnection behave: reconnection is submitted by connd only if a zombie connection is being destroyed and there is a pending reconnection request for the corresponding peer. Also, after a few rejections, reconnection will have a time interval between each attempt. Signed-off-by:
-
Liang Zhen authored
If ib_poll_cq returned +ve without initialising ib_wc::wr_id (bug in driver), then o2iblnd will run into unpredictable situation because ib_wc::wr_id may refer to stale tx/rx pointer in stack. It indicates bug in HCA driver if this happened, ko2iblnd should output console error then close current connection. This patch could also be helpful for LU-5271 Signed-off-by:
Isaac Huang <he.huang@intel.com> Reviewed-by:
-
Doug Oucharek authored
A race condition has been found where connd is cleaning up failed connections, the peer ref counter goes to zero, but we stil have a connecting counter > 0. One possible race is when we are retrying a connection by calling kiblnd_connect_peer() which itself fails and decrements the peer ref counter and gets swapped out before it can decrement the connecting counter. connd swaps in and cleans up the connection where it sees a peer ref counter of 1 and a connecting counter of 1. This will trigger the assert seen in LU-7210 when it decrements the peer counter. The solution: be sure to decrement the connecting counter before decrementing the peer counter in the peer connect failure path. Signed-off-by:
Amir Shehata <amir.shehata@intel.com> Reviewed-by:
-
Liang Zhen authored
refcount taken by cmid is not reliable after kiblnd_connreq_done released the glock because this connection is visible to other threads, another thread can find and close this connection right after kiblnd_connreq_done released the glock, if kiblnd_cm_callback for RDMA_CM_EVENT_DISCONNECTED is called, it can release the connection refcount taken by cmid. It means the connection could be destroyed before kiblnd_connreq_done() finish operations on it. Signed-off-by:
-
Amir Shehata authored
Store map-on-demand and peertx credits in the peer, since the peer is persistent. Also made sure that when assigning the parameters received on the connection to the peer structure through create, that if another peer is added before grabbing the lock we assign these parameters to it as well. Signed-off-by:
-
Jeremy Filizetti authored
This patch adds suppoort for ko2iblnd to have different values for peer_credits and map_on_demand between systems. Signed-off-by:
Jeremy Filizetti <jeremy.filizetti@gmail.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3322 Reviewed-on: http://review.whamcloud.com/11794Reviewed-by:
-
- 02 Mar, 2016 23 commits
-
-
Bhumika Goyal authored
Remove typedefs from structures stt_timer_t and lst_ping_data_t as typedef for a structure type is not preferred. The suffix '_t' and the typedefs are removed using coccinelle. Script 1: //Drop typedefs and '_t' @r1@ type T; @@ typedef struct { ... } T; @script:python c1@ T2; T << r1.T; @@ if T[-2:] =="_t": coccinelle.T2 = T[:-2]; print T else: coccinelle.T2=T; @@ type r1.T; identifier c1.T2; @@ -typedef struct + T2 { ... } -T ; Script 2: //Replacement @@ typedef stt_timer_t; typedef lst_ping_data_t; @@ ( - stt_timer_t + struct stt_timer | - lst_ping_data_t + struct lst_ping_data ) Signed-off-by:Bhumika Goyal <bhumirks@gmail.com> Signed-off-by:
-
Amitoj Kaur Chawla authored
Remove unnecessary test on `rc` variable since it has already been tested previously. Fixes following smatch warning: drivers/staging/lustre/lustre/llite/llite_lib.c:1331 ll_setattr_raw() warn: we tested 'rc' before and it was 'false' Signed-off-by:
Amitoj Kaur Chawla <amitoj1606@gmail.com> Signed-off-by:
-
Amir Shehata authored
With the removal of PMR we no longer require ibh_mrs field to be a array so change it to a simple pointer. Signed-off-by:
Frank Zago <fzago@cray.com> Reviewed-by:
-
Frank Zago authored
Cleanup various typos and grammar errors. Signed-off-by:
James Simmons <uja.ornl@gmail.com> Reviewed-by:
Bob Glossman <bob.glossman@intel.com> Reviewed-by:
-
Frank Zago authored
This reduces the code size by about 1KiB. Signed-off-by:
Patrick Farrell <paf@cray.com> Reviewed-by:
John L. Hammond <john.hammond@intel.com> Reviewed-by:
-
Frank Zago authored
This fixes sparse warnings such as: .../o2iblnd.c:424:1: warning: symbol 'kiblnd_get_peer_info' was not declared. Should it be static? This reduces the code size by 400 bytes. The body of "the_o2iblnd" was moved at the end of the file, to avoid having to declare some static prototypes. Signed-off-by: -
James Simmons authored
LNet is an abstraction built on top of network APIs such as infiniband or TCP/IP layer. Since LNet is dependent on these other layers we should ensure LNet headers should always come after the infiniband header since the infiniband headers can influence the LNet header definitions. Signed-off-by:
Shuichi Ihara <sihara@ddn.com> Reviewed-by:
-
Dmitry Eremin authored
The functions sfw_test_buffers() and sfw_unload_test() from LNet selftest both assume sfw_test_instance_t being passed in is never null. This is corrected here. Signed-off-by:
Dmitry Eremin <dmitry.eremin@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-4629 Reviewed-on: http://review.whamcloud.com/9386Reviewed-by:
-
Dmitry Eremin authored
The function lst_test_add_ioctl is always copying lstio_tes_param from userland even if the user doesn't send this data to LNet selftest. Only consider lstio_tes_param data if lstio_tes_param_len is not zero. Signed-off-by:
-
Dmitry Eremin authored
Pointer 'ni' checked for NULL at line 1569 may be passed to function and may be dereferenced there by passing argument 1 to function 'lnet_ni_notify_locked' at line 1621. Signed-off-by:
-
Dmitry Eremin authored
Null pointer 'best_iface' that comes from line 802 may be dereferenced at line 832. Signed-off-by:
-
James Simmons authored
Currently the socklnd driver doesn't support CPT affinity for its peers. Binding peers to a specific CPT and memory allocated to the NUMA node belonging to the CPT should give a performance boost. Signed-off-by:
Olaf Weber <olaf@sgi.com> Reviewed-by:
-
Jian Yu authored
On some customers' systems, the kernel was compiled with HZ defined to 100, instead of 1000. This improves performance for HPC applications. However, to use these systems with Lustre, customers have to re-build Lustre for the kernel because Lustre directly uses the defined constant HZ. Since kernel 2.6.21, some non-HZ dependent timing APIs become non- inline functions, which can be used in Lustre codes to replace the direct HZ access. These kernel APIs include: jiffies_to_msecs() jiffies_to_usecs() jiffies_to_timespec() msecs_to_jiffies() usecs_to_jiffies() timespec_to_jiffies() And here are some samples of the replacement: HZ -> msecs_to_jiffies(MSEC_PER_SEC) n * HZ -> msecs_to_jiffies(n * MSEC_PER_SEC) HZ / n -> msecs_to_jiffies(MSEC_PER_SEC / n) n / HZ -> jiffies_to_msecs(n) / MSEC_PER_SEC n / HZ * 1000 -> jiffies_to_msecs(n) This patch replaces the direct HZ access in lnet module. Signed-off-by:
Jian Yu <jian.yu@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5443Reviewed-by:
Nathaniel Clark <nathaniel.l.clark@intel.com> Reviewed-by:
-
John L. Hammond authored
In the original code change when libcfs_sock_accept() was made into lnet_sock_accept() a call to set_current_state(TASK_INTERRUPTIBLE) got dropped which was restored. For upstream this is an optimization of calling init_waitqueue_entry() only if accept() return -EAGAIN. Also we can remove setting the task to TASK_RUNNING that is not needed. Signed-off-by:
-
Sebastien Buisson authored
Fix 'copy into fixed size buffer' defects found by Coverity version 6.0.3: Copy into fixed size buffer (STRING_OVERFLOW) The fixed-size string might be overrun by copying without checking the length. Signed-off-by:
Sebastien Buisson <sebastien.buisson@bull.net> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-2074 Reviewed-on: http://review.whamcloud.com/4154Reviewed-by:
-
Alyona Romanenko authored
the offset in hash table is overflowed for no wildcard portal. The offset for no wildcard has been corrected as for wildcard in the LU-1622 Signed-off-by:
Alyona Romanenko <alyona.romanenko@seagate.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7774 Reviewed-on: http://review.whamcloud.com/18422Reviewed-by:
-
Olaf Weber authored
In lnet_ptl_match_delay() we check msg->msg_rx_delayed to see whether the message has been added to the delay queue. But this check is done after lnet_ptl_unlock() and lnet_res_unlock(), and the message can be processed and freed before the check. Replace the check with checking rc against LNET_MATCHMD_NONE, which is how the callers of lnet_ptl_match_delay() know whether the message was added to the delay queue. To make this work we reset rc in the loop when there was no match and the message hasn't been delayed. In addition reorganize the code and add comments to clarify the logic. In lnet_ptl_match_md() a similar msg->msg_rx_delayed is replaced for the same reason. Signed-off-by:
Faccini Bruno <bruno.faccini@intel.com> Reviewed-by:
-
Li Xi authored
When setting TBF rules of jobid, parse_nidrange() prints warning messages. However, this is unnecessary and annoying since paring a TBF rule will always try to parse the jobid like a nid. Signed-off-by:
Li Xi <lixi@ddn.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7647 Reviewed-on: http://review.whamcloud.com/17916Reviewed-by:
Emoly Liu <emoly.liu@intel.com> Reviewed-by:
Bobi Jam <bobijam@hotmail.com> Reviewed-by:
Lai Siyao <lai.siyao@intel.com> Reviewed-by:
-
James Simmons authored
It is consider bad style in the linux kernel to return -1 or a positive number for an error. Instead return the appropriate error codes. Signed-off-by:
-
Amir Shehata authored
Since the # of hops is not a mandatory parameter the LU-6060 patch will cause problems to already existing systems since it changes the behavior by which a route is determined down. To fix this case the # of hops now defaults to LNET_UNDEFINED_HOPS if no hop count is specified. LNET_UNDEFINED_HOPS is defined to ((__u32)-1). When it's printed as %d, it displays as -1. __u32 is used through out the call stack for hop count to explicitly define the size of the hop count and to avoid any sizing issues when passing data to and from the kernel. To keep existing behavior both lnet_compare_routes() and LNetDist() will treat undefined hop count as hop count 1. When executing the logic in lnet_parse_rc_info() there is no longer an assumption that the default hop count is 1. If the hop count is 1 then it must've been explicitly set by the user. Signed-off-by:
-
Liang Zhen authored
When lnet_parse_put calls lnet_ptl_match_md, this function can attach current message on the delayed list if there is no match. It means this message can be taken over and freed by another thread who is posting new MD, then it is not safe for caller of lnet_parse_put to check this message again. This patch fixes this issue by adding a local variable "ready_delay" to store corresponding status of lnet_msg, so lnet doesn't need to check the message again if lnet_ptl_match_md returned MATCH_NONE for it. Signed-off-by:
-
Liang Zhen authored
lnet_route_t::lr_downis is marked as zero even if there is no NI to target network, this is wrong and breaks logic of ARF. This patch fixes this problem. Signed-off-by:
-
Oleg Drokin authored
commit 020ecc6f ("staging: lustre: llite: Remove IS_ERR tests") managed to break one of the converted checks by losing a dereference causing the condition to be wrong and hence leading to a crash later on in case of actual error. IS_ERR_OR_NULL(*inode) got converted into !inode. Signed-off-by:
Oleg Drokin <green@linuxhacker.ru> CC: Amitoj Kaur Chawla <amitoj1606@gmail.com> Signed-off-by:
-
