]> err.no Git - linux-2.6/log
linux-2.6
18 years agoIB/uverbs: Don't serialize with ib_uverbs_idr_mutex
Roland Dreier [Sun, 18 Jun 2006 03:44:49 +0000 (20:44 -0700)]
IB/uverbs: Don't serialize with ib_uverbs_idr_mutex

Currently, all userspace verbs operations that call into the kernel
are serialized by ib_uverbs_idr_mutex.  This can be a scalability
issue for some workloads, especially for devices driven by the ipath
driver, which needs to call into the kernel even for datapath
operations.

Fix this by adding reference counts to the userspace objects, and then
converting ib_uverbs_idr_mutex into a spinlock that only protects the
idrs long enough to take a reference on the object being looked up.
Because remove operations may fail, we have to do a slightly funky
two-step deletion, which is described in the comments at the top of
uverbs_cmd.c.

This also still leaves ib_uverbs_idr_lock as a single lock that is
possibly subject to contention.  However, the lock hold time will only
be a single idr operation, so multiple threads should still be able to
make progress, even if ib_uverbs_idr_lock is being ping-ponged.

Surprisingly, these changes even shrink the object code:

add/remove: 23/5 grow/shrink: 4/21 up/down: 633/-693 (-60)

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Make all device methods truly reentrant
Roland Dreier [Sun, 18 Jun 2006 03:37:41 +0000 (20:37 -0700)]
IB/mthca: Make all device methods truly reentrant

Documentation/infiniband/core_locking.txt says:

  All of the methods in struct ib_device exported by a low-level
  driver must be fully reentrant.  The low-level driver is required to
  perform all synchronization necessary to maintain consistency, even
  if multiple function calls using the same object are run
  simultaneously.

However, mthca's modify_qp, modify_srq and resize_cq methods are
currently not reentrant.  Add a mutex to the QP, SRQ and CQ structures
so that these calls can be properly serialized.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Fix memory leak on modify_qp error paths
Roland Dreier [Sun, 18 Jun 2006 03:37:41 +0000 (20:37 -0700)]
IB/mthca: Fix memory leak on modify_qp error paths

Some error paths after the mthca_alloc_mailbox() call in mthca_modify_qp()
just do a "return -EINVAL" without freeing the mailbox.  Convert these
returns to "goto out" to avoid leaking the mailbox storage.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Factor out common idr code
Roland Dreier [Sun, 18 Jun 2006 03:37:40 +0000 (20:37 -0700)]
IB/uverbs: Factor out common idr code

Factor out common code for adding a userspace object to an idr into a
function idr_add_uobj().  This shrinks both the source and object code:

add/remove: 1/0 grow/shrink: 0/6 up/down: 57/-220 (-163)
function                                     old     new   delta
idr_add_uobj                                   -      57     +57
ib_uverbs_create_ah                          543     512     -31
ib_uverbs_create_srq                         662     630     -32
ib_uverbs_reg_mr                             737     699     -38
ib_uverbs_create_cq                          639     600     -39
ib_uverbs_alloc_pd                           485     446     -39
ib_uverbs_create_qp                         1020     979     -41

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Don't decrement usecnt on error paths
Roland Dreier [Sun, 18 Jun 2006 03:37:40 +0000 (20:37 -0700)]
IB/uverbs: Don't decrement usecnt on error paths

In error paths when destroying an object, uverbs should not decrement
associated objects' usecnt, since ib_dereg_mr(), ib_destroy_qp(),
etc. already do that.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Release lock on error path
Ganapathi CH [Sun, 18 Jun 2006 03:37:40 +0000 (20:37 -0700)]
IB/uverbs: Release lock on error path

If ibdev->alloc_ucontext() fails then ib_uverbs_get_context() does not
unlock file->mutex before returning error.

Signed-off by: Ganapathi CH <cganapathi@novell.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/cm: Use address handle helpers
Sean Hefty [Sun, 18 Jun 2006 03:37:40 +0000 (20:37 -0700)]
IB/cm: Use address handle helpers

Use new ib_init_ah_from_wc() and ib_init_ah_from_path() helper
functions to clean up the IB CM.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/sa: Add ib_init_ah_from_path()
Sean Hefty [Sun, 18 Jun 2006 03:37:39 +0000 (20:37 -0700)]
IB/sa: Add ib_init_ah_from_path()

Add a call to initialize address handle attributes given a path record.
This is used by the CM, and would be useful for users of UD QPs.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Add ib_init_ah_from_wc()
Sean Hefty [Sun, 18 Jun 2006 03:37:39 +0000 (20:37 -0700)]
IB: Add ib_init_ah_from_wc()

Add a function to initialize address handle attributes from a work
completion.  This functionality is duplicated by both verbs and the CM.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/ucm: Get rid of duplicate P_Key parameter
Sean Hefty [Sun, 18 Jun 2006 03:37:39 +0000 (20:37 -0700)]
IB/ucm: Get rid of duplicate P_Key parameter

The P_Key is provided into a SIDR REQ in two places, once as a
parameter, and again in the path record.  Remove the P_Key as a
parameter and always use the one given in the path record.

This change has no practical effect on ABI functionality.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Factor out common request reset code
Ishai Rabinovitz [Sun, 18 Jun 2006 03:37:38 +0000 (20:37 -0700)]
IB/srp: Factor out common request reset code

Misc cleanups in ib_srp:
1) I think that it is more efficient to move the req entries from req_list
   to free_list in srp_reconnect_target (rather than rebuild the free_list).
   (In any case this code is shorter).
2) This allows us to reuse code in srp_reset_device and srp_reconnect_target
   and call a new function srp_reset_req.

Signed-off-by: Ishai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Support SRP rev. 10 targets
Ramachandra K [Sun, 18 Jun 2006 03:37:38 +0000 (20:37 -0700)]
IB/srp: Support SRP rev. 10 targets

There has been a change in the format of port identifiers between
revision 10 of the SRP specification and the current revision 16A.

Revision 10 specifies port identifier format as

  lower 8 bytes :  GUID   upper 8 bytes :  Extension

Whereas revision 16A specifies it as

 lower 8 bytes :  Extension  upper 8 bytes :  GUID

There are older targets (e.g. SilverStorm Virtual Fibre Channel
Bridge) which conform to revision 10 of the SRP specification.

The I/O class of revision 10 is 0xFF00 and the I/O class of revision
16A is 0x0100.

For supporting older targets, this patch:

1) Adds a new optional target creation parameter "io_class". Default
   value of io_class is 0x0100 (i.e. revision 16A)
2) Uses the correct port identifier format for targets with IO class
   of 0xFF00 (i.e. conforming to revision 10)

Signed-off-by: Ramachandra K <rkuchimanchi@silverstorm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years ago[SCSI] srp.h: Add I/O Class values
Ramachandra K [Sun, 18 Jun 2006 03:37:38 +0000 (20:37 -0700)]
[SCSI] srp.h: Add I/O Class values

Add enum values for I/O Class values from rev. 10 and rev. 16a SRP
drafts.  The values are used to detect targets that implement obsolete
revisions of SRP, so that the initiator can use the old format for
port identifier when connecting to them.

Signed-off-by: Ramachandra K <rkuchimanchi@silverstorm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/fmr: Use device's max_map_map_per_fmr attribute in FMR pool.
Or Gerlitz [Sun, 18 Jun 2006 03:37:37 +0000 (20:37 -0700)]
IB/fmr: Use device's max_map_map_per_fmr attribute in FMR pool.

When creating a FMR pool, query the IB device and use the returned
max_map_map_per_fmr attribute as for the max number of FMR remaps. If
the device does not suport querying this attribute, use the original
IB_FMR_MAX_REMAPS (32) default.

Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Fill in max_map_per_fmr device attribute
Or Gerlitz [Sun, 18 Jun 2006 03:37:37 +0000 (20:37 -0700)]
IB/mthca: Fill in max_map_per_fmr device attribute

Report the true max_map_per_fmr value from mthca_query_device(),
taking into account the change in FMR remapping introduced by the
Sinai performance optimization.

Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/ipath: Add client reregister event generation
Roland Dreier [Sun, 18 Jun 2006 03:37:37 +0000 (20:37 -0700)]
IB/ipath: Add client reregister event generation

Generate a client reregister event instead of a LID change event when
client reregister bit is set.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Add client reregister event generation
Leonid Arsh [Sun, 18 Jun 2006 03:37:36 +0000 (20:37 -0700)]
IB/mthca: Add client reregister event generation

Change the mthca snoop of MADs that set PortInfo to check if the SM
has set the client reregister bit, and if it has, generate a client
reregister event.  If the bit is not set, just generate a LID change
event as usual.

Signed-off-by: Leonid Arsh <leonida@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Move struct port_info from ipath to <rdma/ib_smi.h>
Leonid Arsh [Sun, 18 Jun 2006 03:37:36 +0000 (20:37 -0700)]
IB: Move struct port_info from ipath to <rdma/ib_smi.h>

Move ipath's struct port_info into <rdma/ib_smi.h>, so that it can be
used by mthca to implement client reregister support.

Remove the __attribute__((packed)) because all the members of the struct
are naturally aligned anyway.

Signed-off-by: Leonid Arsh <leonida@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Handle client reregister events
Leonid Arsh [Sun, 18 Jun 2006 03:37:36 +0000 (20:37 -0700)]
IPoIB: Handle client reregister events

Handle client reregister events by treating them just like LID or
SM changes -- flush all cached paths and rejoin multicast groups.

Signed-off-by: Leonid Arsh <leonida@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Add client reregister event type
Leonid Arsh [Sun, 18 Jun 2006 03:37:35 +0000 (20:37 -0700)]
IB: Add client reregister event type

Add IB_EVENT_CLIENT_REREGISTER to enum so low-level drivers can
generate "client reregister" events.

Signed-off-by: Leonid Arsh <leonida@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Fix kernel unaligned access on ia64
Jack Morgenstein [Mon, 29 May 2006 16:14:05 +0000 (19:14 +0300)]
IPoIB: Fix kernel unaligned access on ia64

Fix misaligned access faults on ia64: never cast a misaligned
neighbour->ha + 4 pointer to union ib_gid type; pass a void * pointer
instead.  The memcpy was being optimized to use full word accesses
because the compiler thought that union ib_gid is always aligned.

The cast in IPOIB_GID_ARG is safe, since it is fixed to access each
byte separately.

Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Avoid using stale last_send counter when reaping AHs
Roland Dreier [Sun, 18 Jun 2006 03:37:34 +0000 (20:37 -0700)]
IPoIB: Avoid using stale last_send counter when reaping AHs

The comparisons of priv->tx_tail to ah->last_send in ipoib_free_ah()
and ipoib_post_receive() are slightly unsafe, because priv->tx_lock is
not held and hence a stale value of ah->last_send might be used, which
would lead to freeing an AH before the driver was really done with it.
The simple way to fix this is to the optimization of early free from
ipoib_free_ah() and unconditionally queue AHs for reaping, and then
take priv->tx_lock in __ipoib_reap_ah().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mad: Check GID/LID when matching requests
Jack Morgenstein [Sun, 18 Jun 2006 03:37:34 +0000 (20:37 -0700)]
IB/mad: Check GID/LID when matching requests

Check GID/LID for requester side when searching for request which
matches received response.  This is in order to guarantee uniqueness
if the same TID is used when requesting via multiple source LIDs (when
LMC is not zero).  Use ports' cached LMC to perform the check.

Further, do not perform LID check for direct-routed packets, since
the permissive LID makes a proper check impossible.

Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Add caching of ports' LMC
Jack Morgenstein [Sun, 18 Jun 2006 03:37:34 +0000 (20:37 -0700)]
IB: Add caching of ports' LMC

Add an LMC cache to struct ib_device, and add a function
ib_get_cached_lmc() to query the cache.

Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/cm: remove unneeded flush_workqueue
Michael S. Tsirkin [Thu, 25 May 2006 21:51:33 +0000 (14:51 -0700)]
IB/cm: remove unneeded flush_workqueue

destroy_workqueue() already does flush_workqueue().

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
18 years agoIB/ucm: convert semaphore to mutex
Sean Hefty [Thu, 25 May 2006 17:03:23 +0000 (10:03 -0700)]
IB/ucm: convert semaphore to mutex

Convert semaphore in ib_ucm_file to a real mutex.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Get rid of "Target has req_lim 0" messages
Roland Dreier [Sun, 18 Jun 2006 03:37:33 +0000 (20:37 -0700)]
IB/srp: Get rid of "Target has req_lim 0" messages

It's perfectly valid for a connection to an SRP target to have a
request limit of 0, so get rid of the message about it, which can spam
kernel logs even with printk_ratelimit().  Keep a count of such events
in a "zero_req_lim" SCSI host attribute instead, so someone who cares
can look at the statistics.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Handle DREQ events from CM
Ishai Rabinovitz [Sun, 18 Jun 2006 03:37:32 +0000 (20:37 -0700)]
IB/srp: Handle DREQ events from CM

Handle IB_CM_DREQ_ERROR and IB_CM_DREQ_RECEIVED events from the CM,
instead of just printing "Unhandled CM event".  In the case of
DREQ_ERROR, just ignore the event -- a TIMEWAIT_EXIT will be generated
also.  For DREQ_RECEIVED, send a DREP in response to shut the
connection down cleanly.

Signed-off-by: Ishai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Mention RFC numbers in documentation
Roland Dreier [Sun, 18 Jun 2006 03:37:32 +0000 (20:37 -0700)]
IPoIB: Mention RFC numbers in documentation

Now that the IETF has released RFCs covering IPoIB, give the numbers in
the documentation for IPoIB.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Allow sg_tablesize to be adjusted
Vu Pham [Sun, 18 Jun 2006 03:37:32 +0000 (20:37 -0700)]
IB/srp: Allow sg_tablesize to be adjusted

Make the sg_tablesize used by SRP adjustable at module load time via a
module parameter.  Calculate the corresponding IU length required to
support this.

Signed-off-by: Vu Pham <vu@mellanox.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Allow cmd_per_lun to be set per target port
Vu Pham [Sun, 18 Jun 2006 03:37:31 +0000 (20:37 -0700)]
IB/srp: Allow cmd_per_lun to be set per target port

Allow userspace to throttle traffic on a given connection to a target
port by adding "max_cmd_per_lun=xyz" to lower the cmd_per_lun value
set for that scsi_host.

Signed-off-by: Vu Pham <vu@mellanox.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Clean up loop in srp_remove_one()
Ishai Rabinovitz [Sun, 18 Jun 2006 03:37:31 +0000 (20:37 -0700)]
IB/srp: Clean up loop in srp_remove_one()

Interrupts will always be enabled in srp_remove_one(), so
spin_lock_irq() can be used instead of spin_lock_irqsave().
Also, the loop takes target->scsi_host->host_lock, so target->state
can just be set to SRP_TARGET_REMOVED witout testing the old value.

Signed-off-by: Ishai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Make needlessly global ib_mad_cache static
Roland Dreier [Sun, 18 Jun 2006 03:37:31 +0000 (20:37 -0700)]
IB: Make needlessly global ib_mad_cache static

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Change target_mutex to a spinlock
Matthew Wilcox [Sun, 18 Jun 2006 03:37:30 +0000 (20:37 -0700)]
IB/srp: Change target_mutex to a spinlock

The SRP driver never sleeps while holding target_mutex, and it's just
used to protect some simple list operations, so hold times will be
short.  So just convert it to a spinlock, which is smaller and faster.

Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Get rid of unneeded use of list_for_each_entry_safe()
Matthew Wilcox [Sun, 18 Jun 2006 03:37:30 +0000 (20:37 -0700)]
IB/srp: Get rid of unneeded use of list_for_each_entry_safe()

list_for_each_entry_safe() is used in one place where the list isn't
modified.  So just change it to list_for_each_entry().

Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Use SCAN_WILD_CARD from SCSI headers
Matthew Wilcox [Sun, 18 Jun 2006 03:37:30 +0000 (20:37 -0700)]
IB/srp: Use SCAN_WILD_CARD from SCSI headers

SCAN_WILD_CARD is indeed available from <scsi/scsi.h>, which is
already included.  So get rid of private hack.

Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Convert FW commands to use wait_for_completion_timeout()
Roland Dreier [Sun, 18 Jun 2006 03:37:30 +0000 (20:37 -0700)]
IB/mthca: Convert FW commands to use wait_for_completion_timeout()

The kernel has had wait_for_completion_timeout() for a long time now.
mthca should use it to handle FW commands timing out, instead of
implementing the same thing in a much more complicated way by using
wait_for_completion() along with a timer that does complete().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/srp: Use FMRs to map gather/scatter lists
Roland Dreier [Sun, 18 Jun 2006 03:37:29 +0000 (20:37 -0700)]
IB/srp: Use FMRs to map gather/scatter lists

Create an SRP FMR pool on HCAs that support FMRs, and use FMRs to map
gather/scatter lists that have more than one entry into a single
memory region that appears virtually contiguous to the SRP target
(which is the RDMA initiator).

This patch bails out on FMR mapping for SCSI commands where the
gather/scatter list cannot be mapped into a single FMR because there
are sub-page-sized entries in middle of the list.  An unaligned
start or end of the list is OK.

Based on a patch by Vu Pham <vuhuong@mellanox.com>.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Remove dead code
Michael S. Tsirkin [Sun, 18 Jun 2006 03:37:29 +0000 (20:37 -0700)]
IB/mthca: Remove dead code

Kill some dead code in mthca_eq.c

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: IP address based RDMA connection manager
Sean Hefty [Sun, 18 Jun 2006 03:37:29 +0000 (20:37 -0700)]
IB: IP address based RDMA connection manager

Kernel connection management agent over InfiniBand that connects based
on IP addresses.  The agent defines a generic RDMA connection
abstraction to support clients wanting to connect over different RDMA
devices.

The agent also handles RDMA device hotplug events on behalf of clients.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: address translation to map IP toIB addresses (GIDs)
Sean Hefty [Sun, 18 Jun 2006 03:37:28 +0000 (20:37 -0700)]
IB: address translation to map IP toIB addresses (GIDs)

Add an address translation service that maps IP addresses to
InfiniBand GID addresses using IPoIB.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years ago[NET]: Export ip_dev_find()
Sean Hefty [Sun, 18 Jun 2006 03:37:28 +0000 (20:37 -0700)]
[NET]: Export ip_dev_find()

Export ip_dev_find() to allow locating a net_device given an IP address.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/cm: Match connection requests based on private data
Sean Hefty [Sun, 18 Jun 2006 03:37:28 +0000 (20:37 -0700)]
IB/cm: Match connection requests based on private data

Extend matching connection requests to listens in the InfiniBand CM to
include private data checks.

This allows applications to listen on the same service identifier,
with private data directing the request to the appropriate application.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: common handling for marshalling parameters to/from userspace
Sean Hefty [Sun, 18 Jun 2006 03:37:27 +0000 (20:37 -0700)]
IB: common handling for marshalling parameters to/from userspace

Provide common handling for marshalling data between userspace clients
and kernel InfiniBand drivers.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: memfree completion with error FW bug workaround
Michael S. Tsirkin [Tue, 13 Jun 2006 14:19:42 +0000 (17:19 +0300)]
IB/mthca: memfree completion with error FW bug workaround

Memfree firmware is in rare cases reporting WQE index == base - 1 in
receive completion with error, instead of (rq size - 1); base is 0 in
mthca.  Here is a patch to avoid kernel crash and report a correct WR
id in this case.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: restore missing PCI registers after reset
Michael S. Tsirkin [Mon, 12 Jun 2006 13:57:51 +0000 (16:57 +0300)]
IB/mthca: restore missing PCI registers after reset

mthca does not restore the following PCI-X/PCI Express registers after reset:
  PCI-X device: PCI-X command register
  PCI-X bridge: upstream and downstream split transaction registers
  PCI Express : PCI Express device control and link control registers

This causes instability and/or bad performance on systems where one of
these registers is set to a non-default value by BIOS.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoLinux v2.6.17 v2.6.17
Linus Torvalds [Sun, 18 Jun 2006 01:49:35 +0000 (18:49 -0700)]
Linux v2.6.17

Being named "Crazed Snow-Weasel" instills a lot of confidence in this
release, so I'm sure this will be one of the better ones.

18 years ago[PATCH] powerpc: enable CPU_FTR_CI_LARGE_PAGE for cell
Arnd Bergmann [Thu, 15 Jun 2006 13:09:16 +0000 (15:09 +0200)]
[PATCH] powerpc: enable CPU_FTR_CI_LARGE_PAGE for cell

Reflect the fact that the Cell Broadband Engine supports 64k
pages by adding the bit to the CPU features.

Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] powerpc: Fix 64k pages on non-partitioned machines
Arnd Bergmann [Thu, 15 Jun 2006 11:15:44 +0000 (21:15 +1000)]
[PATCH] powerpc: Fix 64k pages on non-partitioned machines

The page size encoding passed to tlbie is incorrect for new-style
large pages.  This fixes it.  This doesn't affect anything on older
machines because mmu_psize_defs[psize].penc (the page size encoding)
is 0 for 4k and 16M pages (the two are distinguished by a separate "is
a large page" bit).

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] arm_timer: remove a racy and obsolete PF_EXITING check
Oleg Nesterov [Thu, 15 Jun 2006 16:12:02 +0000 (20:12 +0400)]
[PATCH] arm_timer: remove a racy and obsolete PF_EXITING check

arm_timer() checks PF_EXITING to prevent BUG_ON(->exit_state)
in run_posix_cpu_timers().

However, for some reason it does so only for CPUCLOCK_PERTHREAD
case (which is imho wrong).

Also, this check is not reliable, PF_EXITING could be set on
another cpu without any locks/barriers just after the check,
so it can't prevent from attaching the timer to the exiting
task.

The previous patch makes this check unneeded.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] run_posix_cpu_timers: remove a bogus BUG_ON()
Oleg Nesterov [Thu, 15 Jun 2006 16:11:43 +0000 (20:11 +0400)]
[PATCH] run_posix_cpu_timers: remove a bogus BUG_ON()

do_exit() clears ->it_##clock##_expires, but nothing prevents
another cpu to attach the timer to exiting process after that.
arm_timer() tries to protect against this race, but the check
is racy.

After exit_notify() does 'write_unlock_irq(&tasklist_lock)' and
before do_exit() calls 'schedule() local timer interrupt can find
tsk->exit_state != 0. If that state was EXIT_DEAD (or another cpu
does sys_wait4) interrupted task has ->signal == NULL.

At this moment exiting task has no pending cpu timers, they were
cleanuped in __exit_signal()->posix_cpu_timers_exit{,_group}(),
so we can just return from irq.

John Stultz recently confirmed this bug, see

http://marc.theaimsgroup.com/?l=linux-kernel&m=115015841413687

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] check_process_timers: fix possible lockup
Oleg Nesterov [Thu, 15 Jun 2006 16:11:15 +0000 (20:11 +0400)]
[PATCH] check_process_timers: fix possible lockup

If the local timer interrupt happens just after do_exit() sets PF_EXITING
(and before it clears ->it_xxx_expires) run_posix_cpu_timers() will call
check_process_timers() with tasklist_lock + ->siglock held and

check_process_timers:

t = tsk;
do {
....

do {
t = next_thread(t);
} while (unlikely(t->flags & PF_EXITING));
} while (t != tsk);

the outer loop will never stop.

Actually, the window is bigger.  Another process can attach the timer
after ->it_xxx_expires was cleared (see the next commit) and the 'if
(PF_EXITING)' check in arm_timer() is racy (see the one after that).

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sky2: netconsole suspend/resume interaction
Stephen Hemminger [Fri, 16 Jun 2006 19:10:46 +0000 (12:10 -0700)]
[PATCH] sky2: netconsole suspend/resume interaction

A couple of fixes that should prevent crashes when using netconsole and
suspend/resume. First, netconsole poll routine shouldn't run unless the
device is up; second, the NAPI poll should be disabled during suspend.

This is only an issue on sky2, because it has to have one NAPI poll
routine for both ports on dual port boards. Normal drivers use
netif_rx_schedule_prep and that checks for netif_running.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix missing ret assignment in __bio_map_user() error path
Jens Axboe [Fri, 16 Jun 2006 11:02:29 +0000 (13:02 +0200)]
[PATCH] Fix missing ret assignment in __bio_map_user() error path

If get_user_pages() returns less pages than what we asked for, we jump
to out_unmap which will return ERR_PTR(ret).  But ret can contain a
positive number just smaller than local_nr_pages, so be sure to set it
to -EFAULT always.

Problem found and diagnosed by Damien Le Moal <damien@sdl.hitachi.co.jp>

Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix cdrom open
Jens Axboe [Fri, 16 Jun 2006 05:46:37 +0000 (07:46 +0200)]
[PATCH] fix cdrom open

Some time ago the cdrom open routine was changed so that we call the
driver's open routine before checking to see if it is read only.  However,
if we discovered that a read write open was not possible and the open
flags required a writable open, we just returned -EROFS without calling
the driver's release routine.   This seems to work for most cdrom drivers,
but breaks the Powerpc iSeries virtual cdrom rather badly.

This just inserts the release call in the error path to balance the call
to "->open()" done by "open_for_data()".

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cfq-iosched: fix crash in do_div()
Jens Axboe [Wed, 14 Jun 2006 17:11:57 +0000 (19:11 +0200)]
[PATCH] cfq-iosched: fix crash in do_div()

We don't clear the seek stat values in cfq_alloc_io_context(), and if
->seek_mean is unlucky enough to be set to -36 by chance, the first
invocation of cfq_update_io_seektime() will oops with a divide by zero
in do_div().

Just memset the entire cic instead of filling invididual values
independently.

Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Return error in case flock_lock_file failure
Kirill Korotaev [Wed, 14 Jun 2006 13:59:35 +0000 (17:59 +0400)]
[PATCH] Return error in case flock_lock_file failure

If flock_lock_file() failed to allocate flock with locks_alloc_lock()
then "error = 0" is returned. Need to return some non-zero.

Signed-off-by: Pavel Emelianov <xemul@openvz.org>
Signed-off-by: Kirill Korotaev <dev@openvz.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sky2: stop/start hardware idle timer on suspend/resume
Stephen Hemminger [Tue, 13 Jun 2006 08:17:31 +0000 (17:17 +0900)]
[PATCH] sky2: stop/start hardware idle timer on suspend/resume

The resume bug was caused not by an early interrupt but because the idle
timeout was not being stopped on suspend.  Also disable hardware IRQ's
on suspend.  Will need to revisit this with hotplug?

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sky2: save/restore base hardware irq during suspend/resume
Stephen Hemminger [Tue, 13 Jun 2006 08:17:30 +0000 (17:17 +0900)]
[PATCH] sky2: save/restore base hardware irq during suspend/resume

The hardware should be fully shut off during suspend, and the base
irq mask restored during resume.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sky2: fix hotplug detect during poll
Stephen Hemminger [Tue, 13 Jun 2006 08:17:29 +0000 (17:17 +0900)]
[PATCH] sky2: fix hotplug detect during poll

If the poll routine detects no hardware available, it needs to dequeue
it self from the network poll list. Linus didn't understand NAPI.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sky2: don't hard code number of ports
Stephen Hemminger [Tue, 13 Jun 2006 08:17:28 +0000 (17:17 +0900)]
[PATCH] sky2: don't hard code number of ports

It is cleaner, to not loop over both ports if only one exists.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sky2: set_power_state should be void
Stephen Hemminger [Tue, 13 Jun 2006 08:17:27 +0000 (17:17 +0900)]
[PATCH] sky2: set_power_state should be void

The set power state function is cleaner if it doesn't return anything.
The only caller that could fail is in suspend() and it can check the argument
there.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] alpha: generic hweight build fix
Randy Dunlap [Mon, 12 Jun 2006 22:13:40 +0000 (15:13 -0700)]
[PATCH] alpha: generic hweight build fix

From: Randy Dunlap <rdunlap@xenotime.net>

According to include/asm-alpha/bitops.h, only ALPHA_EV67 has hardware
hweight support, so ALPHA_EV6 needs to use GENERIC_HWEIGHT.

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Ernst Herzberg <earny@net4u.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] tmpfs: Decrement i_nlink correctly in shmem_rmdir()
Sergey Vlasov [Mon, 12 Jun 2006 20:53:23 +0000 (21:53 +0100)]
[PATCH] tmpfs: Decrement i_nlink correctly in shmem_rmdir()

shmem_rmdir() must undo the increment of i_nlink done in
shmem_get_inode() for directories, otherwise at least
IN_DELETE_SELF inotify event generation is broken.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] tmpfs: time granularity fix for [acm]time going backwards
Robin H. Johnson [Mon, 12 Jun 2006 20:50:25 +0000 (21:50 +0100)]
[PATCH] tmpfs: time granularity fix for [acm]time going backwards

I noticed a strange behavior in a tmpfs file system the other day, while
building packages - occasionally, and seemingly at random, make decided to
rebuild a target. However, only on tmpfs.

A file would be created, and if checked, it had a sub-second timestamp.
However, after an utimes related call where sub-seconds should be set, they
were zeroed instead. In the case that a file was created, and utimes(...,NULL)
was used on it in the same second, the timestamp on the file moved backwards.

After some digging, I found that this was being caused by tmpfs not having a
time granularity set, thus inheriting the default 1 second granularity.

Hugh adds: yes, we missed tmpfs when the s_time_gran mods went into 2.6.11.
Unfortunately, the granularity of CURRENT_TIME, often used in filesystems,
does not match the default granularity set by alloc_super.  A few more such
discrepancies have been found, but this is the most important to fix now.

Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
Acked-by: Andi Kleen <ak@suse.de>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Mon, 12 Jun 2006 20:46:43 +0000 (13:46 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC64]: Do not double-export sys_close() when CONFIG_SOLARIS_EMUL_MODULE

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Mon, 12 Jun 2006 20:46:27 +0000 (13:46 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [IPV4]: Increment ipInHdrErrors when TTL expires.
  [TCP]: continued: reno sacked_out count fix
  [DCCP] Ackvec: fix soft lockup in ackvec handling code

18 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Mon, 12 Jun 2006 20:45:41 +0000 (13:45 -0700)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] Fix Integrator and Versatile interrupt initialisation
  [ARM] 3546/1: PATCH: subtle lost interrupts bug on i.MX
  [ARM] 3547/1: PXA-OHCI: Allow platforms to specify a power budget
  [ARM] Fix Neponset IRQ handling

18 years ago[IPV4]: Increment ipInHdrErrors when TTL expires.
Weidong [Mon, 12 Jun 2006 20:09:59 +0000 (13:09 -0700)]
[IPV4]: Increment ipInHdrErrors when TTL expires.

Signed-off-by: Weidong <weid@nanjing-fnst.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[sky2] Fix sky2 network driver suspend/resume
Linus Torvalds [Mon, 12 Jun 2006 19:53:27 +0000 (12:53 -0700)]
[sky2] Fix sky2 network driver suspend/resume

This fixes two independent problems: it would not save the PCI state on
suspend (and thus try to resume a nonexistent state on resume), and
while shut off, if an interrupt happened on the same shared irq, the irq
handler would react very badly to the interrupt status being an invalid
all-ones state.

Acked-by: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Mon, 12 Jun 2006 14:47:04 +0000 (07:47 -0700)]
Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  [PATCH] sata_mv: grab host lock inside eng_timeout

18 years ago[TCP]: continued: reno sacked_out count fix
Aki M Nyrhinen [Mon, 12 Jun 2006 04:18:56 +0000 (21:18 -0700)]
[TCP]: continued: reno sacked_out count fix

From: Aki M Nyrhinen <anyrhine@cs.helsinki.fi>

IMHO the current fix to the problem (in_flight underflow in reno)
is incorrect.  it treats the symptons but ignores the problem. the
problem is timing out packets other than the head packet when we
don't have sack. i try to explain (sorry if explaining the obvious).

with sack, scanning the retransmit queue for timed out packets is
fine because we know which packets in our retransmit queue have been
acked by the receiver.

without sack, we know only how many packets in our retransmit queue the
receiver has acknowledged, but no idea which packets.

think of a "typical" slow-start overshoot case, where for example
every third packet in a window get lost because a router buffer gets
full.

with sack, we check for timeouts on those every third packet (as the
rest have been sacked). the packet counting works out and if there
is no reordering, we'll retransmit exactly the packets that were
lost.

without sack, however, we check for timeout on every packet and end up
retransmitting consecutive packets in the retransmit queue. in our
slow-start example, 2/3 of those retransmissions are unnecessary. these
unnecessary retransmissions eat the congestion window and evetually
prevent fast recovery from continuing, if enough packets were lost.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[DCCP] Ackvec: fix soft lockup in ackvec handling code
Andrea Bittau [Mon, 12 Jun 2006 03:58:33 +0000 (20:58 -0700)]
[DCCP] Ackvec: fix soft lockup in ackvec handling code

A soft lockup existed in the handling of ack vector records.
Specifically, when a tail of the list of ack vector records was
removed, it was possible to end up iterating infinitely on an element
of the tail.

Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Do not double-export sys_close() when CONFIG_SOLARIS_EMUL_MODULE
David S. Miller [Mon, 12 Jun 2006 04:00:46 +0000 (21:00 -0700)]
[SPARC64]: Do not double-export sys_close() when CONFIG_SOLARIS_EMUL_MODULE

It is already exported by fs/open.c

Noticed by Ben Collins.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] Fix for the PPTP hangs that have been reported
Paul Mackerras [Mon, 12 Jun 2006 02:16:26 +0000 (12:16 +1000)]
[PATCH] Fix for the PPTP hangs that have been reported

People have been reporting that PPP connections over ptys, such as
used with PPTP, will hang randomly when transferring large amounts of
data, for instance in http://bugzilla.kernel.org/show_bug.cgi?id=6530.
I have managed to reproduce the problem, and the patch below fixes the
actual cause.

The problem is not in fact in ppp_async.c but in n_tty.c.  What
happens is that when pptp reads from the pty, we call read_chan() in
drivers/char/n_tty.c on the master side of the pty.  That copies all
the characters out of its buffer to userspace and then calls
check_unthrottle(), which calls the pty unthrottle routine, which
calls tty_wakeup on the slave side, which calls ppp_asynctty_wakeup,
which calls tasklet_schedule.  So far so good.  Since we are in
process context, the tasklet runs immediately and calls
ppp_async_process(), which calls ppp_async_push, which calls the
tty->driver->write function to send some more output.

However, tty->driver->write() returns zero, because the master
tty->receive_room is still zero.  We haven't returned from
check_unthrottle() yet, and read_chan() only updates tty->receive_room
_after_ calling check_unthrottle.  That means that the driver->write
call in ppp_async_process() returns 0.  That would be fine if we were
going to get a subsequent wakeup call, but we aren't (we just had it,
and the buffer is now empty).

The solution is for n_tty.c to update tty->receive_room _before_
calling the driver unthrottle routine.  The patch below does this.
With this patch I was able to transfer a 900MB file over a PPTP
connection (taking about 25 minutes), whereas without the patch the
connection would always stall in under a minute.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sata_mv: grab host lock inside eng_timeout
Mark Lord [Wed, 7 Jun 2006 16:53:29 +0000 (12:53 -0400)]
[PATCH] sata_mv: grab host lock inside eng_timeout

Bug fix:  mv_eng_timeout() calls mv_err_intr() without first grabbing the host lock,
which can lead to all sorts of interesting scenarios.

This whole error-handling portion of sata_mv is nasty (and will get fixed for
the new EH stuff), but for now this patch will help keep it on life-support.

Signed-off-by: Mark Lord <liml@rtr.ca>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6
Linus Torvalds [Sun, 11 Jun 2006 22:28:04 +0000 (15:28 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6:
  [PATCH] PCI: reverse pci config space restore order
  [PATCH] PCI: Improve PCI config space writeback
  [PATCH] PCI: Error handling on PCI device resume
  [PATCH] PCI: fix pciehp compile issue when CONFIG_ACPI is not enabled

18 years ago[PATCH] typo in vmscan.c
Christoph Lameter [Sun, 11 Jun 2006 22:22:26 +0000 (15:22 -0700)]
[PATCH] typo in vmscan.c

From: Christoph Lameter <clameter@sgi.com>

Looks like a comma was left from the conversion from a struct to an
assignment.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PCI: reverse pci config space restore order
Yu, Luming [Tue, 25 Apr 2006 07:00:34 +0000 (00:00 -0700)]
[PATCH] PCI: reverse pci config space restore order

According to Intel ICH spec, there are several rules that Base Address
should be programmed before IOSE  (PCICMD register ) enabled.

For example ICH7:

12.1.3  SATA : the base address register for the bus master register
               should be programmed before this bit is set.

11.1.3:  PCICMD (USB): The base address register for USB should be
                       programmed before this bit is set.
....

To make sure kernel code follow this rule , and prevent unnecessary
confusion. I proposal this patch.

Signed-off-by: Luming Yu <luming.yu@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] PCI: Improve PCI config space writeback
Dave Jones [Wed, 19 Apr 2006 04:06:51 +0000 (21:06 -0700)]
[PATCH] PCI: Improve PCI config space writeback

At least one laptop blew up on resume from suspend with a black screen due
to a lack of this patch.  By only writing back config space that is
different, we minimise the possibility of accidents like this.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] PCI: Error handling on PCI device resume
Jean Delvare [Tue, 18 Apr 2006 12:49:56 +0000 (14:49 +0200)]
[PATCH] PCI: Error handling on PCI device resume

We currently don't handle errors properly when resuming a PCI device:
* In pci_default_resume() we capture the error code returned by
  pci_enable_device() but don't pass it up to the caller.
  Introduced by commit 95a629657dbe28e44a312c47815b3dc3f1ce0970
* In pci_resume_device(), the errors possibly returned by the driver's
  .resume method or by the generic pci_default_resume() function are
  ignored.

This patch fixes both issues.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] PCI: fix pciehp compile issue when CONFIG_ACPI is not enabled
akpm@osdl.org [Sat, 13 May 2006 15:30:52 +0000 (08:30 -0700)]
[PATCH] PCI: fix pciehp compile issue when CONFIG_ACPI is not enabled

Fix build error when CONFIG_ACPI not defined

Signed-off-by: Kristen Carlson Accardi <kristen.c.accardi@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[SPARC]: Migration cost tune up in sparc smp.
Krzysztof Helt [Sun, 11 Jun 2006 05:03:43 +0000 (22:03 -0700)]
[SPARC]: Migration cost tune up in sparc smp.

This patch sets the max_cache_size value required to tune up
scheduler in SMP systems. Otherwise, the calculated
migration_cost is too high and task scheduling may lock up.

Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Set appropriate max_cache_size.
David S. Miller [Sun, 11 Jun 2006 05:02:17 +0000 (22:02 -0700)]
[SPARC64]: Set appropriate max_cache_size.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Sat, 10 Jun 2006 18:03:51 +0000 (11:03 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC64]: Avoid JBUS errors on some Niagara systems.
  [FUSION]: Fix mptspi.c build with CONFIG_PM not set.
  [TG3]: Handle Sun onboard tg3 chips more correctly.
  [SPARC64]: Dump local cpu registers in sun4v_log_error()

18 years ago[PATCH] powerpc: console_initcall ordering issues
Milton Miller [Sat, 10 Jun 2006 16:54:16 +0000 (09:54 -0700)]
[PATCH] powerpc: console_initcall ordering issues

From: Milton Miller <miltonm@bga.com>

The add_preferred_console call in rtas_console.c was not causing the
console to be selected.  It turns out that the add_preferred_console was
being called after the hvc_console driver was registered.  It only works
when it is called before the console driver is registered.

Reorder hvc_console.o after the hvc_console drivers to allow the selection
during console_initcall processing.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] I2O: Bugfixes to get I2O working again
Markus Lidel [Sat, 10 Jun 2006 16:54:14 +0000 (09:54 -0700)]
[PATCH] I2O: Bugfixes to get I2O working again

From: Markus Lidel <Markus.Lidel@shadowconnect.com>

- Fixed locking of struct i2o_exec_wait in Executive-OSM

- Removed LCT Notify in i2o_exec_probe() which caused freeing memory and
  accessing freed memory during first enumeration of I2O devices

- Added missing locking in i2o_exec_lct_notify()

- removed put_device() of I2O controller in i2o_iop_remove() which caused
  the controller structure get freed to early

- Fixed size of mempool in i2o_iop_alloc()

- Fixed access to freed memory in i2o_msg_get()

See http://bugzilla.kernel.org/show_bug.cgi?id=6561

Signed-off-by: Markus Lidel <Markus.Lidel@shadowconnect.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] powernow-k8 crash workaround
Andrew Morton [Sat, 10 Jun 2006 16:54:13 +0000 (09:54 -0700)]
[PATCH] powernow-k8 crash workaround

From: Andrew Morton <akpm@osdl.org>

Work around the oops reported in
http://bugzilla.kernel.org/show_bug.cgi?id=6478.

Thanks to Ralf Hildebrandt <ralf.hildebrandt@charite.de> for testing and
reporting.

Acked-by: Dave Jones <davej@codemonkey.org.uk>
Cc: "Brown, Len" <len.brown@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Further alterations for memory barrier document
David Howells [Sat, 10 Jun 2006 16:54:12 +0000 (09:54 -0700)]
[PATCH] Further alterations for memory barrier document

From: David Howells <dhowells@redhat.com>

Apply some alterations to the memory barrier document that I worked out
with Paul McKenney of IBM, plus some of the alterations suggested by Alan
Stern.

The following changes were made:

 (*) One of the examples given for what can happen with overlapping memory
     barriers was wrong.

 (*) The description of general memory barriers said that a general barrier is
     a combination of a read barrier and a write barrier.  This isn't entirely
     true: it implies both, but is more than a combination of both.

 (*) The first example in the "SMP Barrier Pairing" section was wrong: the
     loads around the read barrier need to touch the memory locations in the
     opposite order to the stores around the write barrier.

 (*) Added a note to make explicit that the loads should be in reverse order to
     the stores.

 (*) Adjusted the diagrams in the "Examples Of Memory Barrier Sequences"
     section to make them clearer.  Added a couple of diagrams to make it more
     clear as to how it could go wrong without the barrier.

 (*) Added a section on memory speculation.

 (*) Dropped any references to memory allocation routines doing memory
     barriers.  They may do sometimes, but it can't be relied on.  This may be
     worthy of further documentation later.

 (*) Made the fact that a LOCK followed by an UNLOCK should not be considered a
     full memory barrier more explicit and gave an example.

Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Sat, 10 Jun 2006 17:59:39 +0000 (10:59 -0700)]
Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  [PATCH] powerpc: Fix cell blade detection
  [PATCH] powerpc: Fix call to ibm,client-architecture-support
  powerpc: Fix machine check problem on 32-bit kernels

18 years ago[ARM] Fix Integrator and Versatile interrupt initialisation
Russell King [Sat, 10 Jun 2006 11:42:12 +0000 (12:42 +0100)]
[ARM] Fix Integrator and Versatile interrupt initialisation

Both Integrator and Versatile were using set_irq_handler() and
enable_irq(), and working around the initialisation of the
chained interrupt, instead of the more correct
set_irq_chained_handler() function.  Fix Integrator and
Versatile to use the right function, and remove these work-arounds.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[SPARC64]: Avoid JBUS errors on some Niagara systems.
David S. Miller [Sat, 10 Jun 2006 08:06:25 +0000 (01:06 -0700)]
[SPARC64]: Avoid JBUS errors on some Niagara systems.

Doing PCI config space accesses to non-present PCI slots
can result in fatal JBUS errors if the PCI config access
hypervisor call is performed on cpus other than the boot
cpu.

PCI config space accesses to present PCI slots works just
fine.

Recursively traverse the OBP device tree under the PCI
controller node and record all present device IDs into
a small hash table.

Avoid the hypervisor call for any PCI config space access
attempt for a device not recorded in the hash table.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[FUSION]: Fix mptspi.c build with CONFIG_PM not set.
Tom "spot" Callaway [Sat, 10 Jun 2006 00:01:48 +0000 (17:01 -0700)]
[FUSION]: Fix mptspi.c build with CONFIG_PM not set.

Signed-off-by: Tom "spot" Callaway <tcallawa@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[TG3]: Handle Sun onboard tg3 chips more correctly.
David S. Miller [Fri, 9 Jun 2006 18:58:36 +0000 (11:58 -0700)]
[TG3]: Handle Sun onboard tg3 chips more correctly.

Get rid of all the SUN_570X logic and instead:

1) Make sure MEMARB_ENABLE is set when we probe the SRAM
   for config information.  If that is off we will get
   timeouts.

2) Always try to sync with the firmware, if there is no
   firmware running do not treat it as an error and instead
   just report it the first time we notice this condition.

3) If there is no valid SRAM signature, assume the device
   is onboard by setting TG3_FLAG_EEPROM_WRITE_PROT.

Update driver version and release date.

With help from Michael Chan and Fabio Massimo Di Nitto.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Dump local cpu registers in sun4v_log_error()
David S. Miller [Wed, 7 Jun 2006 00:37:41 +0000 (17:37 -0700)]
[SPARC64]: Dump local cpu registers in sun4v_log_error()

This makes the debugging information more usable.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] powerpc: Fix cell blade detection
Arnd Bergmann [Wed, 7 Jun 2006 02:04:18 +0000 (12:04 +1000)]
[PATCH] powerpc: Fix cell blade detection

The IBM Cell blade firmware might confuse the kernel to think it's a
pSeries machine. This fixes it for now. With a bit of luck, the firmware
will be updated to avoid that in the future but currently that patch is
needed.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: Fix call to ibm,client-architecture-support
Benjamin Herrenschmidt [Wed, 7 Jun 2006 02:01:32 +0000 (12:01 +1000)]
[PATCH] powerpc: Fix call to ibm,client-architecture-support

The code in prom_init.c calling the firmware
ibm,client-architecture-support method on pSeries has a bug where it
fails to properly pass the instance handle of the firmware object when
trying to call a method. Result ranges from the call doing nothing to
the firmware crashing. (Found by Segher, thanks !)

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agopowerpc: Fix machine check problem on 32-bit kernels
Paul Mackerras [Fri, 9 Jun 2006 03:02:59 +0000 (13:02 +1000)]
powerpc: Fix machine check problem on 32-bit kernels

This fixes a bug found by Dave Jones that means that it is possible
for userspace to provoke a machine check on 32-bit kernels.  This
also fixes a couple of other places where I found similar problems
by inspection.

Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agoMerge branch 'upstream-fixes' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Thu, 8 Jun 2006 22:16:35 +0000 (15:16 -0700)]
Merge branch 'upstream-fixes' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6

* 'upstream-fixes' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6:
  e1000: remove risky prefetch on next_skb->data
  e1000: fix ethtool test irq alloc as "probe"
  [PATCH] bcm43xx: add DMA rx poll workaround to DMA4

18 years ago[PATCH] s390: fix in-user atomic futex operation.
Martin Schwidefsky [Thu, 8 Jun 2006 08:36:20 +0000 (01:36 -0700)]
[PATCH] s390: fix in-user atomic futex operation.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

__futex_atomic_op needs to do an atomic operation in the user address space,
not the kernel address space.  Add the missing sacf 256/sacf 0 to switch to
the secondary mode before doing the compare-and-swap.  In addition add
another fixup for catch specification exceptions if the compare-and-swap
address is not aligned.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>