openwrt/staging/blogic.git
16 years agoIB/ehca: Prevent RDMA-related connection failures on some eHCA2 hardware
Joachim Fenkes [Thu, 17 Jan 2008 14:07:24 +0000 (15:07 +0100)]
IB/ehca: Prevent RDMA-related connection failures on some eHCA2 hardware

Some HW revisions of eHCA2 may cause an RC connection to break if they
received RDMA Reads over that connection before.  This can be
prevented by assuring that, after the first RDMA Read, the QP receives
a new RDMA Read every few million link packets.

Include code into the driver that inserts an empty (size 0) RDMA Read
into the message stream every now and then if the consumer doesn't
post them frequently enough.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ehca: Add "port connection autodetect mode"
Hoang-Nam Nguyen [Thu, 17 Jan 2008 14:05:45 +0000 (15:05 +0100)]
IB/ehca: Add "port connection autodetect mode"

This patch enhances ehca with a capability to "autodetect" the ports
being connected physically. In order to utilize that function the
module option nr_ports must be set to -1 (default is 2 - two
ports). This feature is experimental and will made the default later.

More detail:

If the user connects only one port to the switch, current code requires
  1) port one to be connected and
  2) module option nr_ports=1 to be given.

If autodetect is enabled, ehca will not wait at creation of the GSI QP
for the respective port to become active. Since firmware does not
accept modify_qp() while the port is down at initialization, we need
to cache all calls to modify_qp() for the SMI/GSI QP and just return a
good return code.

When a port is activated and we get a PORT_ACTIVE event, we replay the
cached modify-qp() parms and re-trigger any posted recv WRs. Only then
do we forward the PORT_ACTIVE event to registered clients.

The result of this autodetect patch is that all ports will be
accessible by the users. Depending on their respective cabling only
those ports that are connected properly will become operable. If a
user tries to modify a regular QP of a non-connected port, modify_qp()
will fail. Furthermore, ibv_devinfo should show the port state
accordingly.

Note that this patch primarily improves the loading behaviour of
ehca. If the cable is removed while the driver is operating and
plugged in again, firmware will handle that properly by sending an
appropriate async event.

Signed-off-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ehca: Define array to store SMI/GSI QPs
Hoang-Nam Nguyen [Thu, 17 Jan 2008 14:04:32 +0000 (15:04 +0100)]
IB/ehca: Define array to store SMI/GSI QPs

Signed-off-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ehca: Remove CQ-QP-link before destroying QP in error path of create_qp()
Hoang-Nam Nguyen [Thu, 17 Jan 2008 14:03:55 +0000 (15:03 +0100)]
IB/ehca: Remove CQ-QP-link before destroying QP in error path of create_qp()

Signed-off-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/iser: Add change_queue_depth method
Erez Zilber [Thu, 17 Jan 2008 09:53:17 +0000 (11:53 +0200)]
IB/iser: Add change_queue_depth method

Add a .change_queue_depth handler to the scsi_host_template in the
iSER driver.  iscsi_change_queue_depth was added to iscsi_tcp in order
to solve the problem of queue depth which was too high for some
targets.  It is also applicable for iSER.

Signed-off-by: Erez Zilber <erezz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/iser: Print information about unhandled RDMA CM events
Erez Zilber [Thu, 17 Jan 2008 09:51:58 +0000 (11:51 +0200)]
IB/iser: Print information about unhandled RDMA CM events

Some RDMA CM events are not supported or not handled in iSER.
This patch adds some info (printk) for the user about them.

Signed-off-by: Erez Zilber <erezz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/fmr_pool: ib_fmr_pool_flush() should flush all dirty FMRs
Olaf Kirch [Wed, 16 Jan 2008 17:37:16 +0000 (18:37 +0100)]
IB/fmr_pool: ib_fmr_pool_flush() should flush all dirty FMRs

When a FMR is released via ib_fmr_pool_unmap(), the FMR usually ends
up on the free_list rather than the dirty_list (because we allow a
certain number of remappings before actually requiring a flush).

However, ib_fmr_batch_release() only looks at dirty_list when flushing
out old mappings.  This means that when ib_fmr_pool_flush() is used to
force a flush of the FMR pool, some dirty FMRs that have not reached
their maximum remap count will not actually be flushed.

Fix this by flushing all FMRs that have been used at least once in
ib_fmr_batch_release().

Signed-off-by: Olaf Kirch <olaf.kirch@oracle.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/fmr_pool: Flush serial numbers can get out of sync
Olaf Kirch [Wed, 16 Jan 2008 17:36:27 +0000 (18:36 +0100)]
IB/fmr_pool: Flush serial numbers can get out of sync

Normally, the serial numbers for flush requests and flushes executed
for an FMR pool should be in sync.

However, if the FMR pool flushes dirty FMRs because the
dirty_watermark was reached, we wake up the cleanup thread and let it
do its stuff.  As a side effect, the cleanup thread increments
pool->flush_ser, which leaves it one higher than pool->req_ser.  The
next time the user calls ib_flush_fmr_pool(), the cleanup thread will
be woken up, but ib_flush_fmr_pool() won't wait for the flush to
complete because flush_ser is already past req_ser.  This means the
FMRs that the user expects to be flushed may not have all been flushed
when the function returns.

Fix this by telling the cleanup thread to do work exclusively by
incrementing req_ser, and by moving the comparison of dirty_len and
dirty_watermark into ib_fmr_pool_unmap().

Signed-off-by: Olaf Kirch <olaf.kirch@oracle.com>
16 years agoIB/umad: Simplify and fix locking
Roland Dreier [Fri, 25 Jan 2008 22:15:42 +0000 (14:15 -0800)]
IB/umad: Simplify and fix locking

In addition to being overly complex, the locking in user_mad.c is
broken: there were multiple reports of deadlocks and lockdep warnings.
In particular it seems that a single thread may end up trying to take
the same rwsem for reading more than once, which is explicitly
forbidden in the comments in <linux/rwsem.h>.

To solve this, we change the locking to use plain mutexes instead of
rwsems.  There is one mutex per open file, which protects the contents
of the struct ib_umad_file, including the array of agents and list of
queued packets; and there is one mutex per struct ib_umad_port, which
protects the contents, including the list of open files.  We never
hold the file mutex across calls to functions like ib_unregister_mad_agent(),
which can call back into other ib_umad code to queue a packet, and we
always hold the port mutex as long as we need to make sure that a
device is not hot-unplugged from under us.

This even makes things nicer for users of the -rt patch, since we
remove calls to downgrade_write() (which is not implemented in -rt).

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix some sparse warnings about shadowed symbols
Roland Dreier [Fri, 25 Jan 2008 22:15:42 +0000 (14:15 -0800)]
IB/ipath: Fix some sparse warnings about shadowed symbols

There are a few places in the ipath driver where a variable is
re-declared within a block where it is already in scope.  Most of these
extra declarations can simply be removed, since the variable from the
outer scope is used in a way so that it does not need to keep its
variable across the block with the re-declaration.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRDMA/cxgb3: Endianness annotation for irs field
Roland Dreier [Fri, 25 Jan 2008 22:15:42 +0000 (14:15 -0800)]
RDMA/cxgb3: Endianness annotation for irs field

t3_rdma_init_wr.irs is a big-endian field, so declare it as __be32.
This fixes one sparse warning.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ehca: Use round_jiffies() for EQ polling timer
Anton Blanchard [Mon, 15 Oct 2007 05:49:07 +0000 (00:49 -0500)]
IB/ehca: Use round_jiffies() for EQ polling timer

Use round_jiffies() to align ehca's 1-second timer with other timers
and potentially save power by sleeping cores for longer.

Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRDMA/cma: Override default responder_resources with user value
Sean Hefty [Fri, 4 Jan 2008 18:47:12 +0000 (10:47 -0800)]
RDMA/cma: Override default responder_resources with user value

By default, the responder_resources parameter is set to that received
in a connection request.  The passive side may override this value
when accepting the connection.  Use the value provided by the passive
side when transitioning the QP to RTR state, rather than the value
given in the connect request.  Without this change, the RTR transition
may fail if the passive side supports fewer responder_resources than
that in the request.

For code consistency and to protect against QP destruction, restructure
overriding initiator_depth to match how responder_resources is set.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Drop support for the original QHT7040 board
Dave Olson [Mon, 7 Jan 2008 05:12:38 +0000 (21:12 -0800)]
IB/ipath: Drop support for the original QHT7040 board

The original QHT7040 had significant performance issues so there was an
additional check in the driver for a newer serial number.  Support for
the small quantities of that board shipped has been dropped, so this
patch removes the special checks to simplify the code.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Add ipath_read_ireg() abstraction
Arthur Jones [Mon, 7 Jan 2008 05:12:38 +0000 (21:12 -0800)]
IB/ipath: Add ipath_read_ireg() abstraction

Different chips have different width interrupt status registers, so add
a flag and accessor function to decide which width register read to use.

Signed-off-by: Arthur Jones <arthur.jones@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Add flag and handling for chips with swapped register bug
Ralph Campbell [Mon, 7 Jan 2008 05:12:38 +0000 (21:12 -0800)]
IB/ipath: Add flag and handling for chips with swapped register bug

The 6110 had a bug that caused some registers to be swapped; it was
fixed for the 7220 (and didn't affect the 6120 because it had fewer
registers).  This adds a flag and related code to handle that, and
includes some minor cleanups in the same area.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
16 years agoIB/ipath: Port config has on-chip effects for 7220
Ralph Campbell [Mon, 7 Jan 2008 05:02:34 +0000 (21:02 -0800)]
IB/ipath: Port config has on-chip effects for 7220

The number of configured ports for the 7220 changes the number of eager
TIDs available per port, for all but port 0 (kernel port) which remains
constant, so add a field to give port0 count separate from the portdata
structure.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Allow more flexible user register alignments
Ralph Campbell [Mon, 7 Jan 2008 05:02:34 +0000 (21:02 -0800)]
IB/ipath: Allow more flexible user register alignments

User registers have different alignments on different chips (4KB on
older, 64KB on 7220).  Allow mapping the user registers on kernels with
page sizes up to 64K.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Clean up some comments
Dave Olson [Mon, 7 Jan 2008 05:02:34 +0000 (21:02 -0800)]
IB/ipath: Clean up some comments

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Export hardware counters more consistently
Ralph Campbell [Mon, 7 Jan 2008 05:02:34 +0000 (21:02 -0800)]
IB/ipath: Export hardware counters more consistently

Various hardware counters are exported via the ipath file system (since
it is binary data).  The old file format was very dependent on the HW
offsets for these registers.  Newer HCA chips can have different
counters at different offsets.  This patch adds a level of indirection
to make the file format consistent across HCAs.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: MAD performance sampling registers support
Ralph Campbell [Mon, 7 Jan 2008 05:02:33 +0000 (21:02 -0800)]
IB/ipath: MAD performance sampling registers support

Add support for QLogic HCAs which have hardware performance sampling
registers for PortSamplesControl and PortSamplesResult MADs.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/srp: Add identifying information to log messages
David Dillow [Mon, 7 Jan 2008 23:23:41 +0000 (18:23 -0500)]
IB/srp: Add identifying information to log messages

When you have multiple targets, it gets really confusing when you try
to track down who did a reset when there is no identifying information
in the log message, especially when the same extension ID is mapped
through two different local IB ports.  So, add an identifier that can
be used to track back to which local IB port/remote target pair is the
one having problems.

Signed-off-by: David Dillow <dillowda@ornl.gov>
Acked-by: Pete Wyckoff <pw@osc.edu>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB/CM: Enable SRQ support on HCAs that support fewer than 16 SG entries
Pradeep Satyanarayana [Fri, 21 Dec 2007 21:08:23 +0000 (13:08 -0800)]
IPoIB/CM: Enable SRQ support on HCAs that support fewer than 16 SG entries

Some HCAs (such as ehca2) support SRQ, but only support fewer than 16 SG
entries for SRQs.  Currently IPoIB/CM implicitly assumes all HCAs will
support 16 SG entries for SRQs (to handle a 64K MTU with 4K pages). This
patch removes that restriction by limiting the maximum MTU in connected
mode to what the maximum number of SRQ SG entries allows.

This patch addresses <https://bugs.openfabrics.org/show_bug.cgi?id=728>

Signed-off-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/srp: Enable SG list chaining
David Dillow [Wed, 19 Dec 2007 22:09:15 +0000 (17:09 -0500)]
IB/srp: Enable SG list chaining

By default, the SCSI mid-layer seems to send down 512KB requests
(sg_tablesize = 256), with some requests occasionally combined. By
allowing the mid-layer to chain requests, we can easily grow to 1024KB
or larger -- I've tested 4096KB I/O requests with no problems.

I looked through the DMA paths on the hardware drivers to ensure they
could take advantage of the SG chaining, and it seems that every one
except ipath uses the system's DMA routines, which have been converted
to handle chaining.  ipath looks like it should be OK, but I have no
way to test it.

Signed-off-by: David Dillow <dillowda@ornl.gov>
[ Tested on ipath.  - Roland ]

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/srp: Respect target credit limit
David Dillow [Wed, 19 Dec 2007 22:08:43 +0000 (17:08 -0500)]
IB/srp: Respect target credit limit

The current SRP initiator will send requests even if it has no credits
available.  The results of sending extra requests are vendor specific,
but on some devices, overrunning credits will cost 85% of peak
performance -- e.g. 100 MB/s vs 720 MB/s.  Other devices may just drop
the requests.

This patch will tell the SCSI midlayer to queue requests if there are
fewer than two credits remaining, and will not issue a task management
request if there are no credits remaining.  The mid-layer will retry
the queued command once an outstanding command completes.

The patch also removes the unlikely() in __srp_get_tx_iu(), as it is
not at all unlikely to hit this limit under heavy load.

Signed-off-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB: improve IPv4/IPv6 to IB mcast mapping functions
Rolf Manderscheid [Mon, 10 Dec 2007 20:38:41 +0000 (13:38 -0700)]
IPoIB: improve IPv4/IPv6 to IB mcast mapping functions

An IPoIB subnet on an IB fabric that spans multiple IB subnets can't
use link-local scope in multicast GIDs.  The existing routines that
map IP/IPv6 multicast addresses into IB link-level addresses hard-code
the scope to link-local, and they also leave the partition key field
uninitialised.  This patch adds a parameter (the link-level broadcast
address) to the mapping routines, allowing them to initialise both the
scope and the P_Key appropriately, and fixes up the call sites.

The next step will be to add a way to configure the scope for an IPoIB
interface.

Signed-off-by: Rolf Manderscheid <rvm@obsidianresearch.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Changes for fields moving from devdata to portdata
Dave Olson [Thu, 6 Dec 2007 08:28:02 +0000 (00:28 -0800)]
IB/ipath: Changes for fields moving from devdata to portdata

This patch moves some arrays that were defined per-device to be
variables defined in the per context data structure, thus avoiding extra
kzalloc() calls.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Generalize some xxx_SHIFT macros
Dave Olson [Fri, 21 Dec 2007 09:50:59 +0000 (01:50 -0800)]
IB/ipath: Generalize some xxx_SHIFT macros

In preparation for upcoming chips that have different values for
INFINIPATH_R_PORTENABLE_SHIFT, INFINIPATH_R_INTRAVAIL_SHIFT,
INFINIPATH_R_TAILUPD_SHIFT, and portcfg_shift, remove the shared
#defines and use device-specific variables instead.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: kreceive uses portdata rather than devdata
Ralph Campbell [Thu, 20 Dec 2007 10:43:23 +0000 (02:43 -0800)]
IB/ipath: kreceive uses portdata rather than devdata

kreceive is now portdata * instead of devdata * and other kreceive
related cleanups....

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Cleanup ipath_get_egrbuf()
Ralph Campbell [Sat, 22 Dec 2007 00:47:04 +0000 (16:47 -0800)]
IB/ipath: Cleanup ipath_get_egrbuf()

Remove an unused parameter and fix up the comment.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix RNR NAK handling
Ralph Campbell [Sat, 15 Dec 2007 03:22:34 +0000 (19:22 -0800)]
IB/ipath: Fix RNR NAK handling

This patch fixes a couple of minor problems with RNR NAK handling:
 - The insertion sort was causing extra delay when inserting ahead
   vs. behind an existing entry on the list.
 - A resend of a first packet of a message which is still not ready,
   needs another RNR NAK (i.e., it was suppressed when it shouldn't).
 - Also, the resend tasklet doesn't need to be woken up unless the
   ACK/NAK actually indicates progress has been made.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ehca: Forward event client-reregister-required to registered clients
Hoang-Nam Nguyen [Thu, 20 Dec 2007 14:06:33 +0000 (15:06 +0100)]
IB/ehca: Forward event client-reregister-required to registered clients

This patch allows ehca to forward event client-reregister-required to
registered clients.  One such event is generated by a switch eg. after
its reboot.

Signed-off-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mlx4: Micro-optimize mlx4_ib_poll_one()
Roland Dreier [Fri, 25 Jan 2008 22:15:34 +0000 (14:15 -0800)]
IB/mlx4: Micro-optimize mlx4_ib_poll_one()

Rather than byte-swapping cqe->g_mlpath_rqpn each time we extract a
field from it, byte-swap it once into a temporary variable.  This
results in smaller, better code -- eg, on 32-bit x86:

add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-5 (-5)
function                                     old     new   delta
mlx4_ib_poll_cq                             1188    1183      -5

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mthca: Remove MSI support as scheduled
Adrian Bunk [Tue, 1 Jan 2008 13:47:10 +0000 (15:47 +0200)]
IB/mthca: Remove MSI support as scheduled

Remove MSI support from the mthca driver, as scheduled.  There is no
reason to use MSI instead of MSI-X, since MSI-X performs better.  No
one has spoken up since MSI support was deprecated in commit f6be6fbe
("IB/mthca: Schedule MSI support for removal"), so apparently the MSI
support is unused.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/iser: Typo fix (s/destory/destroy/)
Oliver Pinter [Fri, 25 Jan 2008 22:15:32 +0000 (14:15 -0800)]
IB/iser: Typo fix (s/destory/destroy/)

Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/iser: update URLs of iSER docs
Erez Zilber [Fri, 25 Jan 2008 22:15:32 +0000 (14:15 -0800)]
IB/iser: update URLs of iSER docs

Signed-off-by: Erez Zilber <erezz@voltaire.com>
16 years agoRDMA/cma: add support for rdma_migrate_id()
Sean Hefty [Wed, 14 Nov 2007 08:29:50 +0000 (00:29 -0800)]
RDMA/cma: add support for rdma_migrate_id()

This is based on user feedback from Doug Ledford at RedHat:

Events that occur on an rdma_cm_id are reported to userspace through an
event channel.  Connection request events are reported on the event
channel associated with the listen.  When the connection is accepted, a
new rdma_cm_id is created and automatically uses the listen event
channel.  This is suboptimal where the user only wants listen events on
that channel.

Additionally, it may be desirable to have events related to connection
establishment use a different event channel than those related to
already established connections.

Allow the user to migrate an rdma_cm_id between event channels. All
pending events associated with the rdma_cm_id are moved to the new event
channel.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRDMA/cma: Reenable device removal on passive side
Vladimir Sokolovsky [Sat, 8 Dec 2007 04:32:03 +0000 (20:32 -0800)]
RDMA/cma: Reenable device removal on passive side

Enable conn_id remove on the passive side after connection
establishment.  This corrects an issue where the IB driver can't be
unloaded after running applications over RDS.  The 'dev_remove' counter
does not reach 0 for established connections on the passive side.

This problem is limited to device removal, and only occurs on the
passive side if there are established connections.

Signed-off-by: Vladimir Sokolovsky <vlad@mellanox.co.il>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mad: Fix incorrect access to items on local_list
Sean Hefty [Sat, 1 Dec 2007 01:30:18 +0000 (17:30 -0800)]
IB/mad: Fix incorrect access to items on local_list

In cancel_mads(), MADs are moved from the wait_list and local_list
to a cancel_list for processing.  However, the structures on these two
lists are not the same.  The wait_list references struct
ib_mad_send_wr_private, but local_list references struct
ib_mad_local_private.  Cancel_mads() treats all items moved to the
cancel_list as struct ib_mad_send_wr_private.  This leads to a system
crash when requests are moved from the local_list to the cancel_list.

Fix this by leaving local_list alone.  All requests on the local_list
have completed are just awaiting processing by a queued worker thread.

Bug (crash) reported by Dotan Barak <dotanb@dev.mellanox.co.il>.
Problem with local_list access reported by Robert Reynolds
<rreynolds@opengridcomputing.com>.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/cm: Add basic performance counters
Sean Hefty [Tue, 17 Jul 2007 04:49:35 +0000 (21:49 -0700)]
IB/cm: Add basic performance counters

Add performance/debug counters to track sent/received messages, retries,
and duplicates.  Counters are tracked per CM message type, per port.

The counters are always enabled, so intrusive state tracking is not done.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mad: Report number of times a mad was retried
Sean Hefty [Tue, 27 Nov 2007 08:11:04 +0000 (00:11 -0800)]
IB/mad: Report number of times a mad was retried

To allow ULPs to tune timeout values and capture retry statistics,
report the number of times that a mad send operation was retried.

For RMPP mads, report the total number of times that the any portion
(send window) of the send operation was retried.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/multicast: Report errors on multicast groups if P_key changes
Sean Hefty [Tue, 23 Oct 2007 04:52:54 +0000 (21:52 -0700)]
IB/multicast: Report errors on multicast groups if P_key changes

P_key changes can invalidate multicast groups.  Report errors on all
multicast groups affected by a pkey change.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB: Spelling fixes in comments
Joe Perches [Mon, 17 Dec 2007 19:30:36 +0000 (11:30 -0800)]
IB: Spelling fixes in comments

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agomlx4_core: Fix max_eqs masking in QUERY_DEV_CAP
Jack Morgenstein [Mon, 10 Dec 2007 03:25:23 +0000 (05:25 +0200)]
mlx4_core: Fix max_eqs masking in QUERY_DEV_CAP

log_max_eqs is a 4-bit field, not a 3-bit field in the response to the
QUERY_DEV_CAP FW command, so we should mask with 0xf instead of 0x7
when reading it.

Found by Yossi Leybovitch of Mellanox.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Convert from .nopage to .fault
Nick Piggin [Thu, 13 Dec 2007 23:58:57 +0000 (15:58 -0800)]
IB/ipath: Convert from .nopage to .fault

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Add the work completion error code to the QP error debug output
Ralph Campbell [Sat, 3 Nov 2007 00:40:36 +0000 (17:40 -0700)]
IB/ipath: Add the work completion error code to the QP error debug output

Add the work completion error code to the QP error debug output.
This makes it easier to determine the cause of the error.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Better comment for rmb() in ipath_intr()
Arthur Jones [Thu, 18 Oct 2007 01:18:29 +0000 (18:18 -0700)]
IB/ipath: Better comment for rmb() in ipath_intr()

An internal code review found the comment here lacking -- update it with
more specifics of how and why the rmb() is there.

Signed-off-by: Arthur Jones <arthur.jones@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix comments for ipath_create_srq()
Ralph Campbell [Tue, 27 Nov 2007 07:44:15 +0000 (23:44 -0800)]
IB/ipath: Fix comments for ipath_create_srq()

During a code review, someone noticed the comments didn't match the code.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix error returned from ib_resize_cq if new size smaller than # entries
Ralph Campbell [Fri, 9 Nov 2007 03:53:01 +0000 (19:53 -0800)]
IB/ipath: Fix error returned from ib_resize_cq if new size smaller than # entries

The gen2_basic tests check for the errno value when a CQ is resized
smaller than the number of outstanding completions queue on the CQ.
This patch changes ib_ipath to return EINVAL which is what ib_mthca
returns and what gen2_basic expects.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix sendctrl locking
John Gregor [Wed, 5 Sep 2007 08:57:14 +0000 (01:57 -0700)]
IB/ipath: Fix sendctrl locking

Code review pointed out that the locking around uses of ipath_sendctrl
and kr_sendctrl were, in several places, incorrect and/or inconsistent.

Signed-off-by: John Gregor <john.gregor@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Remove dead code for user process waiting for send buffer
Ralph Campbell [Sat, 13 Oct 2007 03:48:06 +0000 (20:48 -0700)]
IB/ipath: Remove dead code for user process waiting for send buffer

At one point in time there was code to allow a user process to
wait for a send buffer if none were available. This feature was
never used and most of the code was removed. This removes
some missed unused code.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRDMA/cxgb3: Support version 5.0 firmware
Steve Wise [Mon, 26 Nov 2007 17:28:46 +0000 (11:28 -0600)]
RDMA/cxgb3: Support version 5.0 firmware

The 5.0 firmware now supports translating sgls in recv work requests,
so remove the host driver logic currently doing the translation.

Note: this change requires 5.0 firmware.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRDMA/cxgb3: Hold rtnl_lock() around ethtool get_drvinfo call
Steve Wise [Mon, 26 Nov 2007 17:28:44 +0000 (11:28 -0600)]
RDMA/cxgb3: Hold rtnl_lock() around ethtool get_drvinfo call

Currently the call into cxgb3 to get the driver info is not serialized.
The iw_cxgb3 module needs to hold the rtnl_lock around the ethtool ops
call like dev_ioctl() does.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agodrivers/infiniband: Add missing "space"
Joe Perches [Tue, 20 Nov 2007 01:48:11 +0000 (17:48 -0800)]
drivers/infiniband: Add missing "space"

Add missing spaces in the middle of format strings.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Convert ipath_eep_sem semaphore to a mutex
Matthias Kaehlcke [Thu, 15 Nov 2007 23:23:25 +0000 (15:23 -0800)]
IB/ipath: Convert ipath_eep_sem semaphore to a mutex

Signed-off-by: Matthias Kaehlcke <matthias.kaehlcke@gmail.com>
Acked-by: Michael Albaugh <Michael.Albaugh@qlogic.com>
Tested-by: Arthur Jones <arthur.jones@qlogic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mad: Enable loopback of DR SMP responses from userspace
Steve Welch [Tue, 23 Oct 2007 22:06:10 +0000 (15:06 -0700)]
IB/mad: Enable loopback of DR SMP responses from userspace

The local loopback of an outgoing DR SMP response is limited to those
that originate at the driver specific SMA implementation during the
driver specific process_mad() function.  This patch enables a
returning DR SMP originating in userspace (or elsewhere) to be
delivered to the local managment stack.  In this specific case the
driver process_mad() function does not consume or process the MAD, so
a reponse mad has not be created and the original MAD must manually be
copied to the MAD buffer that is to be handed off to the local agent.

Signed-off-by: Steve Welch <swelch@systemfabricworks.com>
Acked-by: Hal Rosenstock <hal@xsigo.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Enable loopback of DR SMP responses from userspace
Ralph Campbell [Tue, 23 Oct 2007 22:07:41 +0000 (15:07 -0700)]
IB/ipath: Enable loopback of DR SMP responses from userspace

This patch is in response to reviewing a patch to the core MAD
processing which fixes loopback of directed route packets to/from user
level MAD agents.  This change enables the core code to work for
ib_ipath by fixing the return code from the ipath process_mad method.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mad: Remove redundant NULL pointer check in ib_mad_recv_done_handler()
Ralph Campbell [Tue, 23 Oct 2007 22:04:15 +0000 (15:04 -0700)]
IB/mad: Remove redundant NULL pointer check in ib_mad_recv_done_handler()

In ib_mad_recv_done_handler(), the response pointer is checked for
NULL after allocating it.  It is then checked again in the local
process_mad() path but there is no possibility of it changing in
between.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Acked-by: Hal Rosenstock <hal@xsigo.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRDMA/iwcm: Set initiator depth and responder resources to device max values
Steve Wise [Mon, 29 Oct 2007 16:34:05 +0000 (11:34 -0500)]
RDMA/iwcm: Set initiator depth and responder resources to device max values

Set the initiator depth and responder resources to the device max
values for new connect request events in the iWARP connection manager.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Improve interrupt handler cache footprint
Dave Olson [Wed, 10 Oct 2007 12:10:35 +0000 (05:10 -0700)]
IB/ipath: Improve interrupt handler cache footprint

Improve interrupt handler cache footprint by noinline'ing error
functions that are rarely called.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB/cm: Add connected mode support for devices without SRQs
Pradeep Satyanarayana [Fri, 25 Jan 2008 22:15:24 +0000 (14:15 -0800)]
IPoIB/cm: Add connected mode support for devices without SRQs

Some IB adapters (notably IBM's eHCA) do not implement SRQs (shared
receive queues).  The current IPoIB connected mode support only works
on devices that support SRQs.

Fix this by adding support for using the receive queue of each
connected mode receive QP.  The disadvantage of this compared to using
an SRQ is that it means a full queue of receives must be posted for
each remote connected mode peer, which means that total memory usage
is potentially much higher than when using SRQs.  To manage this, add
a new module parameter "max_nonsrq_conn_qp" that limits the number of
connections allowed per interface.

The rest of the changes are fairly straightforward: we use a table of
struct ipoib_cm_rx to hold all the active connections, and put the
table index of the connection in the high bits of receive WR IDs.
This is needed because we cannot rely on the struct ib_wc.qp field for
non-SRQ receive completions.  Most of the rest of the changes just
test whether or not an SRQ is available, and post receives or find
received packets in the right place depending on the answer.

Cleaning up dead connections actually becomes simpler, because we do
not have to do the "last WQE reached" dance that is required to
destroy QPs attached to an SRQ.  We just move the QP to the error
state and wait for all pending receives to be flushed.

Signed-off-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
[ Completely rewritten and split up, based on Pradeep's work.  Several
  bugs fixed and no doubt several bugs introduced.  - Roland ]

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB/cm: Factor out ipoib_cm_free_rx_reap_list()
Roland Dreier [Fri, 25 Jan 2008 22:15:24 +0000 (14:15 -0800)]
IPoIB/cm: Factor out ipoib_cm_free_rx_reap_list()

Factor out the code for going through the rx_reap list of struct
ipoib_cm_rx and freeing each one.  This consolidates the code
duplicated between ipoib_cm_dev_stop() and ipoib_cm_rx_reap() and
reduces the risk of error when adding additional accounting.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB/cm: Factor out ipoib_cm_create_srq()
Roland Dreier [Fri, 25 Jan 2008 22:15:24 +0000 (14:15 -0800)]
IPoIB/cm: Factor out ipoib_cm_create_srq()

Factor out the code to create an SRQ and allocate the receive ring in
ipoib_cm_dev_init() into a new function ipoib_cm_create_srq().  This
will make the code neater when support for devices that don't implement
SRQs is added.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB/cm: Factor out ipoib_cm_free_rx_ring()
Roland Dreier [Fri, 25 Jan 2008 22:15:24 +0000 (14:15 -0800)]
IPoIB/cm: Factor out ipoib_cm_free_rx_ring()

Factor out the code to unmap/free skbs and free the receive ring in
ipoib_cm_dev_cleanup() into a new function ipoib_cm_free_rx_ring().
This function will be called from a couple of other places when
support for devices that don't implement SRQs is added.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIPoIB: Trivial formatting cleanups
Roland Dreier [Wed, 24 Oct 2007 02:57:54 +0000 (19:57 -0700)]
IPoIB: Trivial formatting cleanups

Fix whitespace blunders, convert "foo* bar" to "foo *bar", etc.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix crash on unload introduced by sysfs changes
Roland Dreier [Fri, 25 Jan 2008 22:15:21 +0000 (14:15 -0800)]
IB/ipath: Fix crash on unload introduced by sysfs changes

Commit 23b9c1ab ("Infiniband: make ipath driver use default driver
groups.") introduced a bug in the ipath driver where
ipath_device_create_group() fell through into the error path, even on
success, which meant that the sysfs groups it created would always get
removed right away.  This made ipath_device_remove_group() hit the
BUG_ON() in sysfs_remove_group() when it tried to remove those groups a
second time.

Correct the return path so that the groups stick around until they are
supposed to be cleaned up.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Fri, 25 Jan 2008 16:44:29 +0000 (08:44 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/selinux-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/selinux-2.6:
  selinux: make mls_compute_sid always polyinstantiate
  security/selinux: constify function pointer tables and fields
  security: add a secctx_to_secid() hook
  security: call security_file_permission from rw_verify_area
  security: remove security_sb_post_mountroot hook
  Security: remove security.h include from mm.h
  Security: remove security_file_mmap hook sparse-warnings (NULL as 0).
  Security: add get, set, and cloning of superblock security information
  security/selinux: Add missing "space"

16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemoen...
Linus Torvalds [Fri, 25 Jan 2008 16:40:02 +0000 (08:40 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/hskinnemoen/avr32-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemoen/avr32-2.6:
  [AVR32] extint: Set initial irq type to low level
  [AVR32] extint: change set_irq_type() handling
  [AVR32] NMI debugging
  [AVR32] constify function pointer tables
  [AVR32] ATNGW100: Update defconfig
  [AVR32] ATSTK1002: Update defconfig
  [AVR32] Kconfig: Choose daughterboard instead of CPU
  [AVR32] Add support for ATSTK1003 and ATSTK1004
  [AVR32] Clean up external DAC setup code
  [AVR32] ATSTK1000: Move gpio-leds setup to setup.c
  [AVR32] Add support for AT32AP7001 and AT32AP7002
  [AVR32] Provide more CPU information in /proc/cpuinfo and dmesg
  [AVR32] Oprofile support
  [AVR32] Include instrumentation menu
  Disable VGA text console for AVR32 architecture
  [AVR32] Enable debugging only when needed
  ptrace: Call arch_ptrace_attach() when request=PTRACE_TRACEME
  [AVR32] Remove redundant try_to_freeze() call from do_signal()
  [AVR32] Drop GFP_COMP for DMA memory allocations

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
Linus Torvalds [Fri, 25 Jan 2008 16:39:18 +0000 (08:39 -0800)]
Merge git://git./linux/kernel/git/steve/gfs2-2.6-nmw

* git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw: (56 commits)
  [GFS2] Allow journal recovery on read-only mount
  [GFS2] Lockup on error
  [GFS2] Fix page_mkwrite truncation race path
  [GFS2] Fix typo
  [GFS2] Fix write alloc required shortcut calculation
  [GFS2] gfs2_alloc_required performance
  [GFS2] Remove unneeded i_spin
  [GFS2] Reduce inode size by moving i_alloc out of line
  [GFS2] Fix assert in log code
  [GFS2] Fix problems relating to execution of files on GFS2
  [GFS2] Initialize extent_list earlier
  [GFS2] Allow page migration for writeback and ordered pages
  [GFS2] Remove unused variable
  [GFS2] Fix log block mapper
  [GFS2] Minor correction
  [GFS2] Eliminate the no longer needed sd_statfs_mutex
  [GFS2] Incremental patch to fix compiler warning
  [GFS2] Function meta_read optimization
  [GFS2] Only fetch the dinode once in block_map
  [GFS2] Reorganize function gfs2_glmutex_lock
  ...

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Fri, 25 Jan 2008 16:38:25 +0000 (08:38 -0800)]
Merge git://git./linux/kernel/git/herbert/crypto-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (125 commits)
  [CRYPTO] twofish: Merge common glue code
  [CRYPTO] hifn_795x: Fixup container_of() usage
  [CRYPTO] cast6: inline bloat--
  [CRYPTO] api: Set default CRYPTO_MINALIGN to unsigned long long
  [CRYPTO] tcrypt: Make xcbc available as a standalone test
  [CRYPTO] xcbc: Remove bogus hash/cipher test
  [CRYPTO] xcbc: Fix algorithm leak when block size check fails
  [CRYPTO] tcrypt: Zero axbuf in the right function
  [CRYPTO] padlock: Only reset the key once for each CBC and ECB operation
  [CRYPTO] api: Include sched.h for cond_resched in scatterwalk.h
  [CRYPTO] salsa20-asm: Remove unnecessary dependency on CRYPTO_SALSA20
  [CRYPTO] tcrypt: Add select of AEAD
  [CRYPTO] salsa20: Add x86-64 assembly version
  [CRYPTO] salsa20_i586: Salsa20 stream cipher algorithm (i586 version)
  [CRYPTO] gcm: Introduce rfc4106
  [CRYPTO] api: Show async type
  [CRYPTO] chainiv: Avoid lock spinning where possible
  [CRYPTO] seqiv: Add select AEAD in Kconfig
  [CRYPTO] scatterwalk: Handle zero nbytes in scatterwalk_map_and_copy
  [CRYPTO] null: Allow setkey on digest_null
  ...

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-2.6
Linus Torvalds [Fri, 25 Jan 2008 16:34:42 +0000 (08:34 -0800)]
Merge git://git./linux/kernel/git/gregkh/driver-2.6

This can be broken down into these major areas:
 - Documentation updates (language translations and fixes, as
   well as kobject and kset documenatation updates.)
 - major kset/kobject/ktype rework and fixes.  This cleans up the
   kset and kobject and ktype relationship and architecture,
   making sense of things now, and good documenation and samples
   are provided for others to use.  Also the attributes for
   kobjects are much easier to handle now.  This cleaned up a LOT
   of code all through the kernel, making kobjects easier to use
   if you want to.
 - struct bus_type has been reworked to now handle the lifetime
   rules properly, as the kobject is properly dynamic.
 - struct driver has also been reworked, and now the lifetime
   issues are resolved.
 - the block subsystem has been converted to use struct device
   now, and not "raw" kobjects.  This patch has been in the -mm
   tree for over a year now, and finally all the issues are
   worked out with it.  Older distros now properly work with new
   kernels, and no userspace updates are needed at all.
 - nozomi driver is added.  This has also been in -mm for a long
   time, and many people have asked for it to go in.  It is now
   in good enough shape to do so.
 - lots of class_device conversions to use struct device instead.
   The tree is almost all cleaned up now, only SCSI and IB is the
   remaining code to fix up...

* git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-2.6: (196 commits)
  Driver core: coding style fixes
  Kobject: fix coding style issues in kobject c files
  Kobject: fix coding style issues in kobject.h
  Driver core: fix coding style issues in device.h
  spi: use class iteration api
  scsi: use class iteration api
  rtc: use class iteration api
  power supply : use class iteration api
  ieee1394: use class iteration api
  Driver Core: add class iteration api
  Driver core: Cleanup get_device_parent() in device_add() and device_move()
  UIO: constify function pointer tables
  Driver Core: constify the name passed to platform_device_register_simple
  driver core: fix build with SYSFS=n
  sysfs: make SYSFS_DEPRECATED depend on SYSFS
  Driver core: use LIST_HEAD instead of call to INIT_LIST_HEAD in __init
  kobject: add sample code for how to use ksets/ktypes/kobjects
  kobject: add sample code for how to use kobjects in a simple manner.
  kobject: update the kobject/kset documentation
  kobject: remove old, outdated documentation.
  ...

16 years agoslab: fix bootstrap on memoryless node
Pekka Enberg [Fri, 25 Jan 2008 06:20:51 +0000 (08:20 +0200)]
slab: fix bootstrap on memoryless node

If the node we're booting on doesn't have memory, bootstrapping kmalloc()
caches resorts to fallback_alloc() which requires ->nodelists set for all
nodes.  Fix that by calling set_up_list3s() for CACHE_CACHE in
kmem_cache_init().

As kmem_getpages() is called with GFP_THISNODE set, this used to work before
because of breakage in 2.6.22 and before with GFP_THISNODE returning pages from
the wrong node if a node had no memory. So it may have worked accidentally and
in an unsafe manner because the pages would have been associated with the wrong
node which could trigger bug ons and locking troubles.

Tested-by: Mel Gorman <mel@csn.ul.ie>
Tested-by: Olaf Hering <olaf@aepfle.de>
Reviewed-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
[ With additional one-liner by Olaf Hering  - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofix oops on rmmod capidrv
Karsten Keil [Fri, 25 Jan 2008 10:55:28 +0000 (11:55 +0100)]
fix oops on rmmod capidrv

Fix overwriting the stack with the version string
(it is currently 10 bytes + zero) when unloading the
capidrv module. Safeguard against overwriting it
should the version string grow in the future.

Should fix Kernel Bug Tracker Bug 9696.

Signed-off-by: Gerd v. Egidy <gerd.von.egidy@intra2net.com>
Acked-by: Karsten Keil <kkeil@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[GFS2] Allow journal recovery on read-only mount
Abhijith Das [Fri, 18 Jan 2008 20:06:37 +0000 (14:06 -0600)]
[GFS2] Allow journal recovery on read-only mount

This patch allows gfs2 to perform journal recovery even if it is mounted
read-only. Strictly speaking, a read-only mount should not be writing to
the filesystem, but we do this only to perform journal recovery. A
read-only mount will fail if we don't recover the dirty journal. Also,
when gfs2 is used as a root filesystem, it will be mounted read-only
before being mounted read-write during the boot sequence. A failed
read-only mount will panic the machine during bootup.

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Lockup on error
Bob Peterson [Sun, 20 Jan 2008 03:50:24 +0000 (21:50 -0600)]
[GFS2] Lockup on error

I spotted this bug while I was digging around.  Looks like it could cause
a lockup in some rare error condition.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Fix page_mkwrite truncation race path
Steven Whitehouse [Thu, 17 Jan 2008 15:12:03 +0000 (15:12 +0000)]
[GFS2] Fix page_mkwrite truncation race path

There was a bug in the truncation/invalidation race path for
->page_mkwrite for gfs2. It ought to return 0 so that the effect is the
same as if the page was truncated at any of the other points at which
the page_lock is dropped. This will result in the restart of the whole
page fault path. If it was due to a real truncation (as opposed to an
invalidate because we let a glock go) then the ->fault path will pick
that up when it gets called again.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Fix typo
Bob Peterson [Wed, 16 Jan 2008 14:45:39 +0000 (08:45 -0600)]
[GFS2] Fix typo

This patch fixes a minor typo.  Surprisingly, it still compiled.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Fix write alloc required shortcut calculation
Steven Whitehouse [Wed, 16 Jan 2008 14:24:05 +0000 (14:24 +0000)]
[GFS2] Fix write alloc required shortcut calculation

The comparison was being made against the wrong quantity.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] gfs2_alloc_required performance
Bob Peterson [Fri, 11 Jan 2008 19:44:50 +0000 (13:44 -0600)]
[GFS2] gfs2_alloc_required performance

This is a small I/O performance enhancement to gfs2.  (Actually, it is a rework of
an earlier version I got wrong).  The idea here is to check if the write extends
past the last block in the file.  If so, the function can save itself a lot of
time and trouble because it knows an allocate will be required.  Benchmarks like
iozone should see better performance.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Remove unneeded i_spin
Bob Peterson [Fri, 11 Jan 2008 19:31:12 +0000 (13:31 -0600)]
[GFS2] Remove unneeded i_spin

This patch removes a vestigial variable "i_spin" from the gfs2_inode
structure.  This not only saves us memory (>300000 of these in memory
for the oom test) it also saves us time because we don't have to
spend time initializing it (i.e. slightly better performance).

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Reduce inode size by moving i_alloc out of line
Steven Whitehouse [Thu, 10 Jan 2008 15:18:55 +0000 (15:18 +0000)]
[GFS2] Reduce inode size by moving i_alloc out of line

It is possible to reduce the size of GFS2 inodes by taking the i_alloc
structure out of the gfs2_inode. This patch allocates the i_alloc
structure whenever its needed, and frees it afterward. This decreases
the amount of low memory we use at the expense of requiring a memory
allocation for each page or partial page that we write. A quick test
with postmark shows that the overhead is not measurable and I also note
that OCFS2 use the same approach.

In the future I'd like to solve the problem by shrinking down the size
of the members of the i_alloc structure, but for now, this reduces the
immediate problem of using too much low-memory on x86 and doesn't add
too much overhead.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Fix assert in log code
Steven Whitehouse [Thu, 10 Jan 2008 14:49:43 +0000 (14:49 +0000)]
[GFS2] Fix assert in log code

Although the values were all being calculated correctly, there was a
race in the assert due to the way it was using atomic variables. This
changes the value we assert on so that we get the same effect by testing
a different variable. This prevents the assert triggering when it shouldn't.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Fix problems relating to execution of files on GFS2
Steven Whitehouse [Tue, 8 Jan 2008 08:14:30 +0000 (08:14 +0000)]
[GFS2] Fix problems relating to execution of files on GFS2

This patch fixes a couple of problems which affected the execution of files
on GFS2. The first is that there was a corner case where inodes were not
always uptodate at the point at which permissions checks were being carried
out, this was resulting in refusal of execute permission, but only on the
first lookup, subsequent requests worked correctly. The second was a problem
relating to incorrect updating of file sizes which was introduced with the
write_begin/end code for GFS2 a little while back.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Abhijith Das <adas@redhat.com>
16 years ago[GFS2] Initialize extent_list earlier
Bob Peterson [Thu, 3 Jan 2008 15:24:53 +0000 (09:24 -0600)]
[GFS2] Initialize extent_list earlier

Here is a patch for the latest upstream GFS2 code:
The journal extent map needs to be initialized sooner than it
currently is.  Otherwise failed mount attempts (e.g. not enough
journals, etc.) may panic trying to access the uninitialized list.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Allow page migration for writeback and ordered pages
Steven Whitehouse [Thu, 3 Jan 2008 11:31:38 +0000 (11:31 +0000)]
[GFS2] Allow page migration for writeback and ordered pages

To improve performance on NUMA, we use the VM's standard page
migration for writeback and ordered pages. Probably we could
also do the same for journaled data, but that would need a
careful audit of the code, so will be the subject of a later
patch.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Remove unused variable
Steven Whitehouse [Wed, 2 Jan 2008 10:16:56 +0000 (10:16 +0000)]
[GFS2] Remove unused variable

The go_drop_th function is never called or referenced.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Fix log block mapper
Steven Whitehouse [Fri, 14 Dec 2007 14:04:34 +0000 (14:04 +0000)]
[GFS2] Fix log block mapper

A missing offset in the calculation.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Minor correction
Bob Peterson [Wed, 12 Dec 2007 23:52:13 +0000 (17:52 -0600)]
[GFS2] Minor correction

This is a small correction to my previously posted patch1.
It just changes a divide to a shift.  It's faster and doesn't
introduce odd dependencies on 32-bit compiles.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Eliminate the no longer needed sd_statfs_mutex
Bob Peterson [Wed, 12 Dec 2007 17:44:41 +0000 (11:44 -0600)]
[GFS2] Eliminate the no longer needed sd_statfs_mutex

This patch eliminates the unneeded sd_statfs_mutex mutex but preserves
the ordering as discussed.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Incremental patch to fix compiler warning
Bob Peterson [Wed, 12 Dec 2007 15:24:08 +0000 (09:24 -0600)]
[GFS2] Incremental patch to fix compiler warning

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Function meta_read optimization
Bob Peterson [Wed, 12 Dec 2007 01:29:17 +0000 (19:29 -0600)]
[GFS2] Function meta_read optimization

This patch optimizes function gfs2_meta_read.  Basically, gfs2_meta_wait
was being called regardless of whether a disk read was requested.
This just pulls that wait into the if that triggers the read.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Only fetch the dinode once in block_map
Bob Peterson [Wed, 12 Dec 2007 01:16:09 +0000 (19:16 -0600)]
[GFS2] Only fetch the dinode once in block_map

Function gfs2_block_map was often looking up the disk inode twice.
This optimizes it so that only does it once.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Reorganize function gfs2_glmutex_lock
Bob Peterson [Wed, 12 Dec 2007 01:13:54 +0000 (19:13 -0600)]
[GFS2] Reorganize function gfs2_glmutex_lock

This patch optimizes the function gfs2_glmutex_lock.
The basic theory is: Why bother initializing a holder, setting up
wait bits and then waiting on them, if you know the glock can be
yours.  So the holder stuff is placed inside the if checking if the
glock is locked.  This one needs careful scrutiny because changing
anything to do with locking should strike terror into one's heart.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Run through full bitmaps quicker in gfs2_bitfit
Bob Peterson [Wed, 12 Dec 2007 01:00:16 +0000 (19:00 -0600)]
[GFS2] Run through full bitmaps quicker in gfs2_bitfit

I eliminated the passing of an unused parameter into gfs2_bitfit called rgd.

This also changes the gfs2_bitfit code that searches for free (or used) blocks.
Before, the code was trying to check for bytes that indicated 4 blocks in
the undesired state.  The problem is, it was spending more time trying to
do this than it actually was saving.  This version only optimizes the case
where we're looking for free blocks, and it checks a machine word at a time.
So on 32-bit machines, it will check 32-bits (16 blocks) and on 64-bit
machines, it will check 64-bits (32 blocks) at a time.  The compiler
optimizes that quite well and we save some time, especially when running
through full bitmaps (like the bitmaps allocated for the journals).

There's probably a more elegant or optimized way to do this, but I haven't
thought of it yet.  I'm open to suggestions.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Get rid of useless "found" variable in quota.c
Bob Peterson [Wed, 12 Dec 2007 00:51:25 +0000 (18:51 -0600)]
[GFS2] Get rid of useless "found" variable in quota.c

This just eliminates an unused variable from the quota code.
Not likely to be a time saver.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Journal extent mapping
Bob Peterson [Wed, 12 Dec 2007 00:49:21 +0000 (18:49 -0600)]
[GFS2] Journal extent mapping

This patch saves a little time when gfs2 writes to the journals by
keeping a mapping between logical and physical blocks on disk.
That's better than constantly looking up indirect pointers in
buffers, when the journals are several levels of indirection
(which they typically are).

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Remove function gfs2_get_block
Bob Peterson [Mon, 10 Dec 2007 20:13:27 +0000 (14:13 -0600)]
[GFS2] Remove function gfs2_get_block

This patch is just a cleanup.  Function gfs2_get_block() just calls
function gfs2_block_map reversing the last two parameters.  By
reversing the parameters, gfs2_block_map() may be called directly
and function gfs2_get_block may be eliminated altogether.
Since this function is done for every block operation,
this streamlines the code and makes it a little bit more efficient.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] use pid for plock owner for nfs clients
David Teigland [Thu, 6 Dec 2007 15:35:25 +0000 (09:35 -0600)]
[GFS2] use pid for plock owner for nfs clients

The fl_owner is that of lockd when posix locks arrive from nfs
clients, so it can't be used to distinguish between lock holders.
Use fl_pid as owner instead; it's the pid of the process on the
nfs client.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] Remove unused variable
Steven Whitehouse [Fri, 30 Nov 2007 08:17:15 +0000 (08:17 +0000)]
[GFS2] Remove unused variable

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
16 years ago[GFS2] patch to check for recursive lock requests in gfs2_rename code path
Abhijith Das [Thu, 29 Nov 2007 20:13:54 +0000 (14:13 -0600)]
[GFS2] patch to check for recursive lock requests in gfs2_rename code path

A certain scenario in the rename code path triggers a kernel BUG()
because it accidentally does recursive locking The first lock is
requested to unlink an already existing inode (replacing a file) and the
second lock is requested when the destination directory needs to alloc
some space. It is rare that these two
events happen during the same rename call, and even more rare that these
two instances try to lock the same rgrp. It is, however, possible.
https://bugzilla.redhat.com/show_bug.cgi?id=404711

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>