leonidk [Thu, 11 May 2006 08:41:40 +0000 (08:41 +0000)]
[MTHCA] 1. preliminary support for livefish devices (not released yet);
2. (36436) fixed with kernel memory overconsumption on creating/destroying AVs;
3. (36433) bugfix: incorrect error code on create_cq with bad number of cqes;
4. (36427) bugfix: incorrect error code on create_qp with bad number of WRs;
5. fixed PREFAST errors;
6. fixed a bug in the cleanup after an unsuccessful driver start up;
7. fixed a bug in using debug print macros;
leonidk [Mon, 8 May 2006 17:33:12 +0000 (17:33 +0000)]
[MTHCA] 1. bugfix: gid lookup use wrong port number;
2. prints improvement to fit WPP;
3. ib_device_attr dont include specific Mellanox data;
4. added mechanism, providing the work with memfree devices in blocking mode, which requires transition a completion event from KVP to UVP.
The mechanism works as follows:
upon create_cq:
UVP allocates a doorbell record in UARC resident area;
KVP stores the address of this record in the CQ object;
upon completion:
KVP posts event to this record (technically, increments sn - completion sequence number);
upon arming the CQ:
UVP consumes the event (technically, uses sn for numbering the next doorbell)
ftillier [Wed, 3 May 2006 18:51:24 +0000 (18:51 +0000)]
[WSD] Add support for controlling SA timeout/retry settings, as well as
connection establishment behavior when the SA queries timeout from default
of falling back on IPoIB to a hard failure.
[IPoIB] More SM reregister handling
- Avoid attempting to bring the IPoIB port up when a SM change event
is received while the link state is not active
- Add proper state transition before calling ipoib_port_up during SM
change handling.
- Only destroy multicast endpoints when the port goes down, just flush
the AVs from unicast endpoints
This changes the default behavior of the provider to favor latency
at the cost of CPU utilization. This was disabled previously because
of bugs in the WinSock DLL that have been fixed in hotfix KB901481.
[IBAL, IBBUS, MT23108, MTHCA] Make KAL static library
- Move driver initialization from al_driver.c to bus_driver.c
- Fix up INF files that referenced IBAL.
- Kernel ALTS test suite is stale, removed from build.
[WSD] Add perfmon counter object implementation. WSD installation now
registers the WSD provider as a performance counter provider, reporting
I/O and completion statistics to perfmon or any other performance
monitoring tool
[OpenSM] - This change fixes the retrieval of the MCMember records according to
Errata MGTWG3280.
Quoting from MGTWG3280:
SA can be queried for multicast groups by sending a SubnAdmGet() or a
SubnAdmGetTable() request to it using the SA query mechanism (see
15.4.4 Administration Query Subsystem on page 923).
What SA returns in response to a query of multicast groups depends
strongly on whether the request is or is not a trusted request; the
degree of trust affects both the data returned in each attribute and
the set of attributes that are returned. See <ref to C15-0.2.2>.
o15-0.2.5 is made obsolete.
So we need to implement the following descriptive text:
SA can be queried for multicast groups by sending a SubnAdmGet() or a
SubnAdmGetTable() request to it using the SA query mechanism (see
15.4.4 Administration Query Subsystem on page 923). SA will return one
MCMemberRecord per multicast group matching the query, except in
cases where trust is specified as indicated in 15.4.1.2 Access Restrictions
For Other Attributes on page 922; in that case all the MCMemberRecords
associated with the multicast group are returned. The MCMemberRecord
will be returned with the PortGID, ProxyJoin, and the JoinState components
set to 0, except where trust is specified as indicated above, in that
case the actual contents for the above components will be provided.
OpenSM: trivial fix for segfault on invalid port info
A previous patch adding explicit set remote port state DOWN in case the
local one is recycled was not safe. In the case of irresponsive remote
port, it would actually try and traverse a null pointer of the port info.
This simple fix is just a little more careful about it.
p_rem_port was also renamed to p_rem_physp as the pointer actually
points to a physical port...
[MTHCA] Burning support improvements:
1. added support of multiple simult. pciconf devices;
2. removed support for several CR space mappings for one process;
3. PCI bus interface is now taken only once;
4. Fixed bug with handling positive non-zero NTSTATUSes;
5. Removed unused unmap_crspace structure;
6. General cleaning;
[MTHCA] Added bugfixes and changes from Linux Gen2 code.
[MTHCA] (LinuxGen2-rev5521) Add support to IBV_SEND_FENCE in post_send.
[MTHCA] (LinuxGen2-rev5520) Add support to IBV_SEND_FENCE in post_send.
[MTHCA] (LinuxGen2-rev5476) Have mthca's create_srq method return the actual capacity of the SRQ that gets created. Also update comments in <rdma/ib_verbs.h> to clarify that this is what is expected from ib_create_srq().
[MTHCA] (LinuxGen2-rev5293) Generate SQ drained events when requested
[MTHCA] (LinuxGen2-rev5369) When debugging is enabled, the mthca_QUERY_DEV_LIM() firmware command
function prints out some of the device limits that it queries.
However the debugging prints happen before all of the fields are
extracted from the firmware response, so some of the values that get
printed are uninitialized junk. Move the prints to the end of the
function to fix this.
[MTHCA] (LinuxGen2-rev5275) Fixing stuff for FMR, not ported yet
[MTHCA] (LinuxGen2-rev5273) Adding stuff for FMR, not ported yet
[MTHCA] (LinuxGen2-rev5244) Cleaning: Remove useless "extern" from function declarations.
[MTHCA] (LinuxGen2-rev5243) Cleaning: remove might_sleep()
[MTHCA] (LinuxGen2-rev5242) Make functions that never fail return void
[MTHCA] (LinuxGen2-rev5223) We have run into the following problem: if a task receives a signal
while in the process of e.g. destroying a resource (which could be
because the relevant file was closed) mthca could bail out from trying
to take a command interface semaphore without performing the
appropriate command to tell hardware that the resource is being
destroyed.As a result we see messages like
ib_mthca 0000:04:00.0: HW2SW_CQ failed (-4)
[MTHCA] (LinuxGen2-rev5221) There are some cards around that have UAR (user access region) size different from 8 MB. Relax our sanity check to make sure that the PCI BAR is big enough to access the UAR size reported by the device firmware instead.
[MTHCA] (LinuxGen2-rev5220) Fix typo in overflow test in mthca receive posting.
tzachid [Wed, 29 Mar 2006 17:14:21 +0000 (17:14 +0000)]
[MTHCA] 1. Applied - with a little improvement - a patch, removing dependency of MTCHA driver from IBAL.SYS.
2. Applied a patch for support of user-mode calls without a UVP.
3. Applied a patch, eliminating usage of mt_time.h and its abstraction.
4. Cleaning: Removed most of the '#ifdef WIN_TO_BE_CHANGED' stuff.
5. Cleaning: Removed '#if 0'.
tzachid [Tue, 28 Mar 2006 09:23:02 +0000 (09:23 +0000)]
[MTHCA] 1. another implementation of atomic functions, which works around MS compiler bug in InterlockedOr (mt_atomic.h).
2. fix of small bugs in debug macro (mlnx_uvp_debug.h).
3. changes for using of WPP in release version.
4. compiler warning fixes (for /W4).
eitan [Mon, 27 Mar 2006 12:10:47 +0000 (12:10 +0000)]
[OpenSM] - fix opts file - comment out all options. The user can undo the commenting out if he wants to use one of the options. Otherwise - opensm will run with its default opts.
tzachid [Sun, 26 Mar 2006 13:13:48 +0000 (13:13 +0000)]
[MTHCA] (continues patch 251) remaps RDD/EEC WC status values to IB_WCS_GENERIC_ERR, and that removes the use of the qp_num field in the ib_wc_t structure
eitan [Sun, 26 Mar 2006 09:42:49 +0000 (09:42 +0000)]
[OpenSM] -
1. ib_types.h: Add defines for min/max MTU/rate
2. Disregard subn->ca_min_mtu and subn->ca_min_rate when new MC group is
created and exact MTU and/or rate values are specified.
projects / ~shefty / rdma-win.git / log