~shefty/rdma-dev.git
12 years agoMerge branch 'kmemleak' of git://linux-arm.org/linux-2.6
Linus Torvalds [Fri, 11 Sep 2009 16:16:22 +0000 (09:16 -0700)]
Merge branch 'kmemleak' of git://linux-arm.org/linux-2.6

* 'kmemleak' of git://linux-arm.org/linux-2.6:
  kmemleak: Improve the "Early log buffer exceeded" error message
  kmemleak: fix sparse warning for static declarations
  kmemleak: fix sparse warning over overshadowed flags
  kmemleak: move common painting code together
  kmemleak: add clear command support
  kmemleak: use bool for true/false questions
  kmemleak: Do no create the clean-up thread during kmemleak_disable()
  kmemleak: Scan all thread stacks
  kmemleak: Don't scan uninitialized memory when kmemcheck is enabled
  kmemleak: Ignore the aperture memory hole on x86_64
  kmemleak: Printing of the objects hex dump
  kmemleak: Do not report alloc_bootmem blocks as leaks
  kmemleak: Save the stack trace for early allocations
  kmemleak: Mark the early log buffer as __initdata
  kmemleak: Dump object information on request
  kmemleak: Allow rescheduling during an object scanning

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Fri, 11 Sep 2009 15:58:32 +0000 (08:58 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (48 commits)
  RDMA/iwcm: Reject the connection when the cm_id is destroyed
  RDMA/cxgb3: Clean up properly on FW mismatch failures
  RDMA/cxgb3: Don't ignore insert_handle() failures
  MAINTAINERS: InfiniBand/RDMA mailing list transition to vger
  IB/mad: Allow tuning of QP0 and QP1 sizes
  IB/mad: Fix possible lock-lock-timer deadlock
  RDMA/nes: Map MTU to IB_MTU_* and correctly report link state
  RDMA/nes: Rework the disconn routine for terminate and flushing
  RDMA/nes: Use the flush code to fill in cqe error
  RDMA/nes: Make poll_cq return correct number of wqes during flush
  RDMA/nes: Use flush mechanism to set status for wqe in error
  RDMA/nes: Implement Terminate Packet
  RDMA/nes: Add CQ error handling
  RDMA/nes: Clean out CQ completions when QP is destroyed
  RDMA/nes: Change memory allocation for cqp request to GFP_ATOMIC
  RDMA/nes: Allocate work item for disconnect event handling
  RDMA/nes: Update refcnt during disconnect
  IB/mthca: Don't allow userspace open while recovering from catastrophic error
  IB/mthca: Distinguish multiple devices in /proc/interrupts
  IB/mthca: Annotate CQ locking
  ...

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Fri, 11 Sep 2009 15:55:49 +0000 (08:55 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/security-testing-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/security-testing-2.6: (57 commits)
  binfmt_elf: fix PT_INTERP bss handling
  TPM: Fixup boot probe timeout for tpm_tis driver
  sysfs: Add labeling support for sysfs
  LSM/SELinux: inode_{get,set,notify}secctx hooks to access LSM security context information.
  VFS: Factor out part of vfs_setxattr so it can be called from the SELinux hook for inode_setsecctx.
  KEYS: Add missing linux/tracehook.h #inclusions
  KEYS: Fix default security_session_to_parent()
  Security/SELinux: includecheck fix kernel/sysctl.c
  KEYS: security_cred_alloc_blank() should return int under all circumstances
  IMA: open new file for read
  KEYS: Add a keyctl to install a process's session keyring on its parent [try #6]
  KEYS: Extend TIF_NOTIFY_RESUME to (almost) all architectures [try #6]
  KEYS: Do some whitespace cleanups [try #6]
  KEYS: Make /proc/keys use keyid not numread as file position [try #6]
  KEYS: Add garbage collection for dead, revoked and expired keys. [try #6]
  KEYS: Flag dead keys to induce EKEYREVOKED [try #6]
  KEYS: Allow keyctl_revoke() on keys that have SETATTR but not WRITE perm [try #6]
  KEYS: Deal with dead-type keys appropriately [try #6]
  CRED: Add some configurable debugging [try #6]
  selinux: Support for the new TUN LSM hooks
  ...

12 years agokmemleak: Improve the "Early log buffer exceeded" error message
Catalin Marinas [Fri, 11 Sep 2009 09:42:09 +0000 (10:42 +0100)]
kmemleak: Improve the "Early log buffer exceeded" error message

Based on a suggestion from Jaswinder, clarify what the user would need
to do to avoid this error message from kmemleak.

Reported-by: Jaswinder Singh Rajput <jaswinder@kernel.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agoMerge branch 'mad' into for-linus
Roland Dreier [Fri, 11 Sep 2009 04:19:45 +0000 (21:19 -0700)]
Merge branch 'mad' into for-linus

Conflicts:
drivers/infiniband/core/mad.c

12 years agoMerge branches 'cxgb3', 'ehca', 'ipath', 'ipoib', 'misc', 'mlx4', 'mthca' and 'nes...
Roland Dreier [Fri, 11 Sep 2009 04:18:07 +0000 (21:18 -0700)]
Merge branches 'cxgb3', 'ehca', 'ipath', 'ipoib', 'misc', 'mlx4', 'mthca' and 'nes' into for-linus

12 years agoMerge branch 'next' into for-linus
James Morris [Thu, 10 Sep 2009 22:04:49 +0000 (08:04 +1000)]
Merge branch 'next' into for-linus

12 years agomd: Fix "strchr" [drivers/md/dm-log-userspace.ko] undefined!
Geert Uytterhoeven [Thu, 10 Sep 2009 21:13:28 +0000 (23:13 +0200)]
md: Fix "strchr" [drivers/md/dm-log-userspace.ko] undefined!

Commit b8313b6da7e2e7c7f47d93d8561969a3ff9ba0ea ("dm log: remove incorrect
field from userspace table output") added a call to strstr() with a
single-character "needle" string parameter.

Unfortunately some versions of gcc replace such calls to strstr() by calls
to strchr() behind our back.  This causes linking errors if strchr() is
defined as an inline function in <asm/string.h> (e.g. on m68k):

| WARNING: "strchr" [drivers/md/dm-log-userspace.ko] undefined!

Avoid this by explicitly calling strchr() instead.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agobinfmt_elf: fix PT_INTERP bss handling
Roland McGrath [Wed, 9 Sep 2009 02:49:40 +0000 (19:49 -0700)]
binfmt_elf: fix PT_INTERP bss handling

In fs/binfmt_elf.c, load_elf_interp() calls padzero() for .bss even if
the PT_LOAD has no PROT_WRITE and no .bss.  This generates EFAULT.

Here is a small test case.  (Yes, there are other, useful PT_INTERP
which have only .text and no .data/.bss.)

----- ptinterp.S
_start: .globl _start
 nop
 int3
-----
$ gcc -m32 -nostartfiles -nostdlib -o ptinterp ptinterp.S
$ gcc -m32 -Wl,--dynamic-linker=ptinterp -o hello hello.c
$ ./hello
Segmentation fault  # during execve() itself

After applying the patch:
$ ./hello
Trace trap  # user-mode execution after execve() finishes

If the ELF headers are actually self-inconsistent, then dying is fine.
But having no PROT_WRITE segment is perfectly normal and correct if
there is no segment with p_memsz > p_filesz (i.e. bss).  John Reiser
suggested checking for PROT_WRITE in the bss logic.  I think it makes
most sense to simply apply the bss logic only when there is bss.

This patch looks less trivial than it is due to some reindentation.
It just moves the "if (last_bss > elf_bss) {" test up to include the
partial-page bss logic as well as the more-pages bss logic.

Reported-by: John Reiser <jreiser@bitwagon.com>
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoTPM: Fixup boot probe timeout for tpm_tis driver
Jason Gunthorpe [Wed, 9 Sep 2009 23:22:18 +0000 (17:22 -0600)]
TPM: Fixup boot probe timeout for tpm_tis driver

When probing the device in tpm_tis_init the call request_locality
uses timeout_a, which wasn't being initalized until after
request_locality. This results in request_locality falsely timing
out if the chip is still starting. Move the initialization to before
request_locality.

This probably only matters for embedded cases (ie mine), a BIOS likely
gets the TPM into a state where this code path isn't necessary.

Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Acked-by: Rajiv Andrade <srajiv@linux.vnet.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoMerge branch 'lookup-permissions-cleanup'
Linus Torvalds [Thu, 10 Sep 2009 03:04:54 +0000 (20:04 -0700)]
Merge branch 'lookup-permissions-cleanup'

* lookup-permissions-cleanup:
  jffs2/jfs/xfs: switch over to 'check_acl' rather than 'permission()'
  ext[234]: move over to 'check_acl' permission model
  shmfs: use 'check_acl' instead of 'permission'
  Make 'check_acl()' a first-class filesystem op
  Simplify exec_permission_lite(), part 3
  Simplify exec_permission_lite() further
  Simplify exec_permission_lite() logic
  Do not call 'ima_path_check()' for each path component

12 years agobinfmt_elf: fix PT_INTERP bss handling
Roland McGrath [Wed, 9 Sep 2009 02:49:40 +0000 (19:49 -0700)]
binfmt_elf: fix PT_INTERP bss handling

In fs/binfmt_elf.c, load_elf_interp() calls padzero() for .bss even if
the PT_LOAD has no PROT_WRITE and no .bss.  This generates EFAULT.

Here is a small test case.  (Yes, there are other, useful PT_INTERP
which have only .text and no .data/.bss.)

----- ptinterp.S
_start: .globl _start
 nop
 int3
-----
$ gcc -m32 -nostartfiles -nostdlib -o ptinterp ptinterp.S
$ gcc -m32 -Wl,--dynamic-linker=ptinterp -o hello hello.c
$ ./hello
Segmentation fault  # during execve() itself

After applying the patch:
$ ./hello
Trace trap  # user-mode execution after execve() finishes

If the ELF headers are actually self-inconsistent, then dying is fine.
But having no PROT_WRITE segment is perfectly normal and correct if
there is no segment with p_memsz > p_filesz (i.e. bss).  John Reiser
suggested checking for PROT_WRITE in the bss logic.  I think it makes
most sense to simply apply the bss logic only when there is bss.

This patch looks less trivial than it is due to some reindentation.
It just moves the "if (last_bss > elf_bss) {" test up to include the
partial-page bss logic as well as the more-pages bss logic.

Reported-by: John Reiser <jreiser@bitwagon.com>
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agosysfs: Add labeling support for sysfs
David P. Quigley [Wed, 9 Sep 2009 18:25:37 +0000 (14:25 -0400)]
sysfs: Add labeling support for sysfs

This patch adds a setxattr handler to the file, directory, and symlink
inode_operations structures for sysfs. The patch uses hooks introduced in the
previous patch to handle the getting and setting of security information for
the sysfs inodes. As was suggested by Eric Biederman the struct iattr in the
sysfs_dirent structure has been replaced by a structure which contains the
iattr, secdata and secdata length to allow the changes to persist in the event
that the inode representing the sysfs_dirent is evicted. Because sysfs only
stores this information when a change is made all the optional data is moved
into one dynamically allocated field.

This patch addresses an issue where SELinux was denying virtd access to the PCI
configuration entries in sysfs. The lack of setxattr handlers for sysfs
required that a single label be assigned to all entries in sysfs. Granting virtd
access to every entry in sysfs is not an acceptable solution so fine grained
labeling of sysfs is required such that individual entries can be labeled
appropriately.

[sds:  Fixed compile-time warnings, coding style, and setting of inode security init flags.]

Signed-off-by: David P. Quigley <dpquigl@tycho.nsa.gov>
Signed-off-by: Stephen D. Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoLSM/SELinux: inode_{get,set,notify}secctx hooks to access LSM security context inform...
David P. Quigley [Thu, 3 Sep 2009 18:25:57 +0000 (14:25 -0400)]
LSM/SELinux: inode_{get,set,notify}secctx hooks to access LSM security context information.

This patch introduces three new hooks. The inode_getsecctx hook is used to get
all relevant information from an LSM about an inode. The inode_setsecctx is
used to set both the in-core and on-disk state for the inode based on a context
derived from inode_getsecctx.The final hook inode_notifysecctx will notify the
LSM of a change for the in-core state of the inode in question. These hooks are
for use in the labeled NFS code and addresses concerns of how to set security
on an inode in a multi-xattr LSM. For historical reasons Stephen Smalley's
explanation of the reason for these hooks is pasted below.

Quote Stephen Smalley

inode_setsecctx:  Change the security context of an inode.  Updates the
in core security context managed by the security module and invokes the
fs code as needed (via __vfs_setxattr_noperm) to update any backing
xattrs that represent the context.  Example usage:  NFS server invokes
this hook to change the security context in its incore inode and on the
backing file system to a value provided by the client on a SETATTR
operation.

inode_notifysecctx:  Notify the security module of what the security
context of an inode should be.  Initializes the incore security context
managed by the security module for this inode.  Example usage:  NFS
client invokes this hook to initialize the security context in its
incore inode to the value provided by the server for the file when the
server returned the file's attributes to the client.

Signed-off-by: David P. Quigley <dpquigl@tycho.nsa.gov>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoVFS: Factor out part of vfs_setxattr so it can be called from the SELinux hook for...
David P. Quigley [Thu, 3 Sep 2009 18:25:56 +0000 (14:25 -0400)]
VFS: Factor out part of vfs_setxattr so it can be called from the SELinux hook for inode_setsecctx.

This factors out the part of the vfs_setxattr function that performs the
setting of the xattr and its notification. This is needed so the SELinux
implementation of inode_setsecctx can handle the setting of the xattr while
maintaining the proper separation of layers.

Signed-off-by: David P. Quigley <dpquigl@tycho.nsa.gov>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoLinux 2.6.31 v2.6.31
Linus Torvalds [Wed, 9 Sep 2009 22:13:59 +0000 (15:13 -0700)]
Linux 2.6.31

12 years agoRDMA/iwcm: Reject the connection when the cm_id is destroyed
Steve Wise [Wed, 9 Sep 2009 18:37:38 +0000 (11:37 -0700)]
RDMA/iwcm: Reject the connection when the cm_id is destroyed

If the cm_id of a connect request is destroyed prior to the ULP
accepting or rejecting the connection, then the provider never cleans
up the connection.  The iwcm should explicitly reject these
connections if the cm_id is destroyed.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: Clean up properly on FW mismatch failures
Steve Wise [Wed, 9 Sep 2009 18:25:56 +0000 (11:25 -0700)]
RDMA/cxgb3: Clean up properly on FW mismatch failures

FW mismatches can cause a crash in the iw_cxgb3 event handler.

- NULL the t3cdev->ulp pointer on failures in cxio_rdev_open()
- Silently ignore events when the ulp ptr is NULL in iwch_err_handler()

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: Don't ignore insert_handle() failures
Steve Wise [Wed, 9 Sep 2009 18:25:55 +0000 (11:25 -0700)]
RDMA/cxgb3: Don't ignore insert_handle() failures

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoaoe: allocate unused request_queue for sysfs
Ed Cashin [Wed, 9 Sep 2009 12:10:18 +0000 (14:10 +0200)]
aoe: allocate unused request_queue for sysfs

Andy Whitcroft reported an oops in aoe triggered by use of an
incorrectly initialised request_queue object:

  [ 2645.959090] kobject '<NULL>' (ffff880059ca22c0): tried to add
an uninitialized object, something is seriously wrong.
  [ 2645.959104] Pid: 6, comm: events/0 Not tainted 2.6.31-5-generic #24-Ubuntu
  [ 2645.959107] Call Trace:
  [ 2645.959139] [<ffffffff8126ca2f>] kobject_add+0x5f/0x70
  [ 2645.959151] [<ffffffff8125b4ab>] blk_register_queue+0x8b/0xf0
  [ 2645.959155] [<ffffffff8126043f>] add_disk+0x8f/0x160
  [ 2645.959161] [<ffffffffa01673c4>] aoeblk_gdalloc+0x164/0x1c0 [aoe]

The request queue of an aoe device is not used but can be allocated in
code that does not sleep.

Bruno bisected this regression down to

  cd43e26f071524647e660706b784ebcbefbd2e44

  block: Expose stacked device queues in sysfs

"This seems to generate /sys/block/$device/queue and its contents for
 everyone who is using queues, not just for those queues that have a
 non-NULL queue->request_fn."

Addresses http://bugs.launchpad.net/bugs/410198
Addresses http://bugzilla.kernel.org/show_bug.cgi?id=13942

Note that embedding a queue inside another object has always been
an illegal construct, since the queues are reference counted and
must persist until the last reference is dropped. So aoe was
always buggy in this respect (Jens).

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Bruno Premont <bonbons@linux-vserver.org>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
12 years agoKEYS: Add missing linux/tracehook.h #inclusions
David Howells [Wed, 9 Sep 2009 07:30:21 +0000 (08:30 +0100)]
KEYS: Add missing linux/tracehook.h #inclusions

Add #inclusions of linux/tracehook.h to those arch files that had the tracehook
call for TIF_NOTIFY_RESUME added when support for that flag was added to that
arch.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoi915: disable interrupts before tearing down GEM state
Linus Torvalds [Wed, 9 Sep 2009 00:09:24 +0000 (17:09 -0700)]
i915: disable interrupts before tearing down GEM state

Reinette Chatre reports a frozen system (with blinking keyboard LEDs)
when switching from graphics mode to the text console, or when
suspending (which does the same thing). With netconsole, the oops
turned out to be

BUG: unable to handle kernel NULL pointer dereference at 0000000000000084
IP: [<ffffffffa03ecaab>] i915_driver_irq_handler+0x26b/0xd20 [i915]

and it's due to the i915_gem.c code doing drm_irq_uninstall() after
having done i915_gem_idle(). And the i915_gem_idle() path will do

  i915_gem_idle() ->
    i915_gem_cleanup_ringbuffer() ->
      i915_gem_cleanup_hws() ->
        dev_priv->hw_status_page = NULL;

but if an i915 interrupt comes in after this stage, it may want to
access that hw_status_page, and gets the above NULL pointer dereference.

And since the NULL pointer dereference happens from within an interrupt,
and with the screen still in graphics mode, the common end result is
simply a silently hung machine.

Fix it by simply uninstalling the irq handler before idling rather than
after. Fixes

    http://bugzilla.kernel.org/show_bug.cgi?id=13819

Reported-and-tested-by: Reinette Chatre <reinette.chatre@intel.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agojffs2/jfs/xfs: switch over to 'check_acl' rather than 'permission()'
Linus Torvalds [Fri, 28 Aug 2009 19:29:03 +0000 (12:29 -0700)]
jffs2/jfs/xfs: switch over to 'check_acl' rather than 'permission()'

This avoids an indirect call in the VFS for each path component lookup.

Well, at least as long as you own the directory in question, and the ACL
check is unnecessary.

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoext[234]: move over to 'check_acl' permission model
Linus Torvalds [Fri, 28 Aug 2009 19:12:24 +0000 (12:12 -0700)]
ext[234]: move over to 'check_acl' permission model

Don't implement per-filesystem 'extX_permission()' functions that have
to be called for every path component operation, and instead just expose
the actual ACL checking so that the VFS layer can now do it for us.

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoshmfs: use 'check_acl' instead of 'permission'
Linus Torvalds [Fri, 28 Aug 2009 19:04:28 +0000 (12:04 -0700)]
shmfs: use 'check_acl' instead of 'permission'

shmfs wants purely standard POSIX ACL semantics, so we can use the new
generic VFS layer POSIX ACL checking rather than cooking our own
'permission()' function.

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMake 'check_acl()' a first-class filesystem op
Linus Torvalds [Fri, 28 Aug 2009 18:51:25 +0000 (11:51 -0700)]
Make 'check_acl()' a first-class filesystem op

This is stage one in flattening out the callchains for the common
permission testing.  Rather than have most filesystem implement their
own inode->i_op->permission function that just calls back down to the
VFS layers 'generic_permission()' with the per-filesystem ACL checking
function, the filesystem can just expose its 'check_acl' function
directly, and let the VFS layer do everything for it.

This is all just preparatory - no filesystem actually enables this yet.

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoSimplify exec_permission_lite(), part 3
Linus Torvalds [Fri, 28 Aug 2009 18:08:31 +0000 (11:08 -0700)]
Simplify exec_permission_lite(), part 3

Don't call down to the generic inode_permission() function just to
call the inode-specific permission function - just do it directly.

The generic inode_permission() code does things like checking MAY_WRITE
and devcgroup_inode_permission(), neither of which are relevant for the
light pathname walk permission checks (we always do just MAY_EXEC, and
the inode is never a special device).

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoSimplify exec_permission_lite() further
Linus Torvalds [Fri, 28 Aug 2009 17:53:56 +0000 (10:53 -0700)]
Simplify exec_permission_lite() further

This function is only called for path components that are already known
to be directories (they have a '->lookup' method).  So don't bother
doing that whole S_ISDIR() testing, the whole point of the 'lite()'
version is that we know that we are looking at a directory component,
and that we're only checking name lookup permission.

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoSimplify exec_permission_lite() logic
Linus Torvalds [Fri, 28 Aug 2009 17:50:37 +0000 (10:50 -0700)]
Simplify exec_permission_lite() logic

Instead of returning EAGAIN and having the caller do something
special for that case,  just do the special case directly.

Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoDo not call 'ima_path_check()' for each path component
Linus Torvalds [Fri, 28 Aug 2009 17:05:33 +0000 (10:05 -0700)]
Do not call 'ima_path_check()' for each path component

Not only is that a supremely timing-critical path, but it's hopefully
some day going to be lockless for the common case, and ima can't do
that.

Plus the integrity code doesn't even care about non-regular files, so it
was always a total waste of time and effort.

Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Mimi Zohar <zohar@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agodrm/i915: fix mask bits setting
Zhenyu Wang [Tue, 8 Sep 2009 06:52:25 +0000 (14:52 +0800)]
drm/i915: fix mask bits setting

eDP is exclusive connector too, and add missing crtc_mask
setting for TV.

This fixes

http://bugzilla.kernel.org/show_bug.cgi?id=14139

Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Reported-and-tested-by: Carlos R. Mafra <crmafra2@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokmemleak: fix sparse warning for static declarations
Luis R. Rodriguez [Tue, 8 Sep 2009 16:31:45 +0000 (17:31 +0100)]
kmemleak: fix sparse warning for static declarations

This fixes these sparse warnings:

mm/kmemleak.c:1179:6: warning: symbol 'start_scan_thread' was not declared. Should it be static?
mm/kmemleak.c:1194:6: warning: symbol 'stop_scan_thread' was not declared. Should it be static?

Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agokmemleak: fix sparse warning over overshadowed flags
Luis R. Rodriguez [Tue, 8 Sep 2009 16:32:34 +0000 (17:32 +0100)]
kmemleak: fix sparse warning over overshadowed flags

A secondary irq_save is not required as a locking before it was
already disabling irqs.

This fixes this sparse warning:
mm/kmemleak.c:512:31: warning: symbol 'flags' shadows an earlier one
mm/kmemleak.c:448:23: originally declared here

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agokmemleak: move common painting code together
Luis R. Rodriguez [Sat, 5 Sep 2009 00:44:52 +0000 (17:44 -0700)]
kmemleak: move common painting code together

When painting grey or black we do the same thing, bring
this together into a helper and identify coloring grey or
black explicitly with defines. This makes this a little
easier to read.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agokmemleak: add clear command support
Luis R. Rodriguez [Sat, 5 Sep 2009 00:44:51 +0000 (17:44 -0700)]
kmemleak: add clear command support

In an ideal world your kmemleak output will be small, when its
not (usually during initial bootup) you can use the clear command
to ingore previously reported and unreferenced kmemleak objects. We
do this by painting all currently reported unreferenced objects grey.
We paint them grey instead of black to allow future scans on the same
objects as such objects could still potentially reference newly
allocated objects in the future.

To test a critical section on demand with a clean
/sys/kernel/debug/kmemleak you can do:

echo clear > /sys/kernel/debug/kmemleak
        test your kernel or modules
echo scan > /sys/kernel/debug/kmemleak

Then as usual to get your report with:

cat /sys/kernel/debug/kmemleak

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agokmemleak: use bool for true/false questions
Luis R. Rodriguez [Tue, 8 Sep 2009 15:34:50 +0000 (16:34 +0100)]
kmemleak: use bool for true/false questions

Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agokmemleak: Do no create the clean-up thread during kmemleak_disable()
Catalin Marinas [Mon, 7 Sep 2009 09:14:42 +0000 (10:14 +0100)]
kmemleak: Do no create the clean-up thread during kmemleak_disable()

The kmemleak_disable() function could be called from various contexts
including IRQ. It creates a clean-up thread but the kthread_create()
function has restrictions on which contexts it can be called from,
mainly because of the kthread_create_lock. The patch changes the
kmemleak clean-up thread to a workqueue.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Reported-by: Eric Paris <eparis@redhat.com>
12 years agoMAINTAINERS: InfiniBand/RDMA mailing list transition to vger
Roland Dreier [Tue, 8 Sep 2009 04:54:38 +0000 (21:54 -0700)]
MAINTAINERS: InfiniBand/RDMA mailing list transition to vger

InfiniBand/RDMA development discussion is moving from
general@lists.openfabrics.org to linux-rdma@vger.kernel.org.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoMerge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Mon, 7 Sep 2009 18:42:25 +0000 (11:42 -0700)]
Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/radeon/kms: add LTE/GTE discard + rv515 two sided stencil register.

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Mon, 7 Sep 2009 18:41:15 +0000 (11:41 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/security-testing-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/security-testing-2.6:
  IMA: update ima_counts_put

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Mon, 7 Sep 2009 18:40:24 +0000 (11:40 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  gianfar: Fix build.

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/ide-2.6
Linus Torvalds [Mon, 7 Sep 2009 18:40:15 +0000 (11:40 -0700)]
Merge git://git./linux/kernel/git/davem/ide-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/ide-2.6:
  pcmcia: add CNF-CDROM-ID for ide

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/anholt...
Linus Torvalds [Mon, 7 Sep 2009 18:38:30 +0000 (11:38 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/anholt/drm-intel

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/anholt/drm-intel:
  agp/intel: support for new chip variant of IGDNG mobile
  drm/i915: Unref old_obj on get_fence_reg() error path
  drm/i915: increase default latency constant (v2 w/comment)

12 years agoIB/mad: Allow tuning of QP0 and QP1 sizes
Hal Rosenstock [Mon, 7 Sep 2009 15:28:48 +0000 (08:28 -0700)]
IB/mad: Allow tuning of QP0 and QP1 sizes

MADs are UD and can be dropped if there are no receives posted, so
allow receive queue size to be set with a module parameter in case the
queue needs to be lengthened.  Send side tuning is done for symmetry
with receive.

Signed-off-by: Hal Rosenstock <hal.rosenstock@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mad: Fix possible lock-lock-timer deadlock
Roland Dreier [Mon, 7 Sep 2009 15:27:50 +0000 (08:27 -0700)]
IB/mad: Fix possible lock-lock-timer deadlock

Lockdep reported a possible deadlock with cm_id_priv->lock,
mad_agent_priv->lock and mad_agent_priv->timed_work.timer; this
happens because the mad module does

cancel_delayed_work(&mad_agent_priv->timed_work);

while holding mad_agent_priv->lock.  cancel_delayed_work() internally
does del_timer_sync(&mad_agent_priv->timed_work.timer).

This can turn into a deadlock because mad_agent_priv->lock is taken
inside cm_id_priv->lock, so we can get the following set of contexts
that deadlock each other:

 A: holding cm_id_priv->lock, waiting for mad_agent_priv->lock
 B: holding mad_agent_priv->lock, waiting for del_timer_sync()
 C: interrupt during mad_agent_priv->timed_work.timer that takes
    cm_id_priv->lock

Fix this by using the new __cancel_delayed_work() interface (which
internally does del_timer() instead of del_timer_sync()) in all the
places where we are holding a lock.

Addresses: http://bugzilla.kernel.org/show_bug.cgi?id=13757
Reported-by: Bart Van Assche <bart.vanassche@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoKEYS: Fix default security_session_to_parent()
David Howells [Mon, 7 Sep 2009 12:24:17 +0000 (13:24 +0100)]
KEYS: Fix default security_session_to_parent()

Fix the default security_session_to_parent() in linux/security.h to have a
body.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agodrm/radeon/kms: add LTE/GTE discard + rv515 two sided stencil register.
Dave Airlie [Mon, 7 Sep 2009 05:26:19 +0000 (15:26 +1000)]
drm/radeon/kms: add LTE/GTE discard + rv515 two sided stencil register.

This adds some rv350+ register for LTE/GTE discard,
and enables the rv515 two sided stencil register.
It also disables the DEPTHXY_OFFSET register which
can be used to workaround the CS checker.
Moves rs690 to proper place in rs600 and uses correct
table on rs600.

Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agoIMA: update ima_counts_put
Mimi Zohar [Fri, 4 Sep 2009 17:08:46 +0000 (13:08 -0400)]
IMA: update ima_counts_put

- As ima_counts_put() may be called after the inode has been freed,
verify that the inode is not NULL, before dereferencing it.

- Maintain the IMA file counters in may_open() properly, decrementing
any counter increments on subsequent errors.

Reported-by: Ciprian Docan <docan@eden.rutgers.edu>
Reported-by: J.R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Mimi Zohar <zohar@us.ibm.com>
Acked-by: Eric Paris <eparis@redhat.com
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoSecurity/SELinux: includecheck fix kernel/sysctl.c
Jaswinder Singh Rajput [Fri, 4 Sep 2009 17:43:21 +0000 (23:13 +0530)]
Security/SELinux: includecheck fix kernel/sysctl.c

fix the following 'make includecheck' warning:

  kernel/sysctl.c: linux/security.h is included more than once.

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agoKEYS: security_cred_alloc_blank() should return int under all circumstances
David Howells [Fri, 4 Sep 2009 08:19:48 +0000 (09:19 +0100)]
KEYS: security_cred_alloc_blank() should return int under all circumstances

Make security_cred_alloc_blank() return int, not void, when CONFIG_SECURITY=n.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
12 years agogianfar: Fix build.
David S. Miller [Sun, 6 Sep 2009 08:41:02 +0000 (01:41 -0700)]
gianfar: Fix build.

Reported by Michael Guntsche <mike@it-loops.com>

--------------------
Commit
38bddf04bcfe661fbdab94888c3b72c32f6873b3 gianfar: gfar_remove needs to call unregister_netdev()

breaks the build of the gianfar driver because "dev" is undefined in
this function. To quickly test rc9 I changed this to priv->ndev but I do
not know if this is the correct one.
--------------------

Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoRDMA/nes: Map MTU to IB_MTU_* and correctly report link state
Chien Tung [Sun, 6 Sep 2009 03:36:39 +0000 (20:36 -0700)]
RDMA/nes: Map MTU to IB_MTU_* and correctly report link state

Old query_port code reports static MTU and link state values.
Instead, map actual MTU to next largest IB_MTU_* constant and
correctly report link state.

Cc: Steve Wise <swise@opengridcomputing.com>
Reported-by: Jeff Squyres <jsquyres@cisco.com>
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Rework the disconn routine for terminate and flushing
Don Wood [Sun, 6 Sep 2009 03:36:39 +0000 (20:36 -0700)]
RDMA/nes: Rework the disconn routine for terminate and flushing

The disconn routine has been reworked to acoomodate the terminate and
flushing changes.  The routine has been reorganized to make all the
decisions at the start then it performs all the required operations.
This simplified the lock handling and is easier to follow.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Use the flush code to fill in cqe error
Don Wood [Sun, 6 Sep 2009 03:36:39 +0000 (20:36 -0700)]
RDMA/nes: Use the flush code to fill in cqe error

Use the flush status to fill in cqe status when a specific error has
been identified.  Subsequent flushed completions still use the flushed
value.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Make poll_cq return correct number of wqes during flush
Don Wood [Sun, 6 Sep 2009 03:36:39 +0000 (20:36 -0700)]
RDMA/nes: Make poll_cq return correct number of wqes during flush

When a flush request is given to the hw, it will place one cqe marked
as flushed (unless there is nothing to flush).  An application that is
waiting for all wqe's to complete will be left hanging.  This modifies
poll_cq to return the correct number of flushes for the pending
elements on the wq.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Use flush mechanism to set status for wqe in error
Don Wood [Sun, 6 Sep 2009 03:36:38 +0000 (20:36 -0700)]
RDMA/nes: Use flush mechanism to set status for wqe in error

When an asynchronous event occurs that requires a terminate, it is
sometimes possible to identify the wqe in error.  This change uses
flush to get this information to the poll routine.  The flush
operation puts the status into the cqe.  If this information is not
available, it continues to use the more generic flush code as before.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Implement Terminate Packet
Don Wood [Sun, 6 Sep 2009 03:36:38 +0000 (20:36 -0700)]
RDMA/nes: Implement Terminate Packet

Implement the sending and receiving of Terminate packets.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Add CQ error handling
Don Wood [Sun, 6 Sep 2009 03:36:38 +0000 (20:36 -0700)]
RDMA/nes: Add CQ error handling

CQ errors are not being handled correctly.  Put in the the upcall for
CQ errors.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Clean out CQ completions when QP is destroyed
Don Wood [Sun, 6 Sep 2009 03:36:37 +0000 (20:36 -0700)]
RDMA/nes: Clean out CQ completions when QP is destroyed

When a QP is destroyed, unprocessed CQ entries could still reference
the QP.  This change zeroes the context value at QP destroy time.  By
skipping over cqe's with a zero context, poll_cq no longer processes a
cqe for a destroyed QP.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Change memory allocation for cqp request to GFP_ATOMIC
Don Wood [Sun, 6 Sep 2009 03:36:37 +0000 (20:36 -0700)]
RDMA/nes: Change memory allocation for cqp request to GFP_ATOMIC

The routine to allocate a cqp request is not called from process
context code.  Since it is not OK to sleep, it needs to use GFP_ATOMIC
not GFP_KERNEL.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Allocate work item for disconnect event handling
Don Wood [Sun, 6 Sep 2009 03:36:37 +0000 (20:36 -0700)]
RDMA/nes: Allocate work item for disconnect event handling

The code currently has a work structure in the QP.  This requires a
lock and a pending flag to ensure there is never more than one request
active.  When two events happen quickly (such as FIN and LLP CLOSE),
it causes unnecessary timeouts since the second one is dropped.

This fix allocates memory for the work request so the second one can
be queued.  A lock is removed since it is no longer needed.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/nes: Update refcnt during disconnect
Don Wood [Sun, 6 Sep 2009 03:36:36 +0000 (20:36 -0700)]
RDMA/nes: Update refcnt during disconnect

During termination, it is possible for the refcnt to go to zero while
the worker thread is posting events upward.  This fix increments the
refcnt before the request is passed to the worker thread.  The thread
decrements the refcnt when the request is completed.

Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mthca: Don't allow userspace open while recovering from catastrophic error
Jack Morgenstein [Sun, 6 Sep 2009 03:36:16 +0000 (20:36 -0700)]
IB/mthca: Don't allow userspace open while recovering from catastrophic error

Userspace apps are supposed to release all ib device resources if they
receive a fatal async event (IBV_EVENT_DEVICE_FATAL).  However, the
app has no way of knowing when the device has come back up, except to
repeatedly attempt ibv_open_device() until it succeeds.

However, currently there is no protection against the open succeeding
while the device is in being removed following the fatal event.  In
this case, the open will succeed, but as a result the device waits in
the middle of its removal until the new app releases its resources --
and the new app will not do so, since the open succeeded at a point
following the fatal event generation.

This patch adds an "active" flag to the device. The active flag is set
to false (in the fatal event flow) before the "fatal" event is
generated, so any subsequent ibv_dev_open() call to the device will
fail until the device comes back up, thus preventing the above
deadlock.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mthca: Distinguish multiple devices in /proc/interrupts
Arputham Benjamin [Sun, 6 Sep 2009 03:36:15 +0000 (20:36 -0700)]
IB/mthca: Distinguish multiple devices in /proc/interrupts

When the mthca driver uses the same name for interrupts for every
device in the system.  This can make it very confusing trying to work
out exactly which device MSI-X interrupts are for.  Change the driver
to add the PCI name of the device to the interrupt name.

Signed-off-by: Arputham Benjamin <abenjamin@sgi.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mthca: Annotate CQ locking
Roland Dreier [Sun, 6 Sep 2009 03:36:15 +0000 (20:36 -0700)]
IB/mthca: Annotate CQ locking

mthca_ib_lock_cqs()/mthca_ib_unlock_cqs() are helper functions that
lock/unlock both CQs attached to a QP in the proper order to avoid
AB-BA deadlocks.  Annotate this so sparse can understand what's going
on (and warn us if we misuse these functions).

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mthca: Remove unnecessary include of <linux/init.h>
Roland Dreier [Sun, 6 Sep 2009 03:36:15 +0000 (20:36 -0700)]
IB/mthca: Remove unnecessary include of <linux/init.h>

mthca_reset.c doesn't have any function annotations, so there's no
reason to include <linux/init.h>.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mthca: Remove unnecessary include of <asm/page.h>
Roland Dreier [Sun, 6 Sep 2009 03:36:13 +0000 (20:36 -0700)]
IB/mthca: Remove unnecessary include of <asm/page.h>

mthca_config_reg.h was including <asm/page.h> for no reason -- the whole
file is just defines of constants, so it's entirely self-contained.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mlx4: Don't allow userspace open while recovering from catastrophic error
Jack Morgenstein [Sun, 6 Sep 2009 03:24:50 +0000 (20:24 -0700)]
IB/mlx4: Don't allow userspace open while recovering from catastrophic error

Userspace apps are supposed to release all ib device resources if they
receive a fatal async event (IBV_EVENT_DEVICE_FATAL).  However, the
app has no way of knowing when the device has come back up, except to
repeatedly attempt ibv_open_device() until it succeeds.

However, currently there is no protection against the open succeeding
while the device is in being removed following the fatal event.  In
this case, the open will succeed, but as a result the device waits in
the middle of its removal until the new app releases its resources --
and the new app will not do so, since the open succeeded at a point
following the fatal event generation.

This patch adds an "active" flag to the device. The active flag is set
to false (in the fatal event flow) before the "fatal" event is
generated, so any subsequent ibv_dev_open() call to the device will
fail until the device comes back up, thus preventing the above
deadlock.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agomlx4_core: Distinguish multiple devices in /proc/interrupts
Arputham Benjamin [Sun, 6 Sep 2009 03:24:50 +0000 (20:24 -0700)]
mlx4_core: Distinguish multiple devices in /proc/interrupts

When the mlx4 driver uses the same name for interrupts for every
device in the system.  This can make it very confusing trying to work
out exactly which device MSI-X interrupts are for.  Change the driver
to add the PCI name of the device to the interrupt name.

Signed-off-by: Arputham Benjamin <abenjamin@sgi.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agomlx4_core: Avoid double free_icms
Yevgeny Petrilin [Sun, 6 Sep 2009 03:24:49 +0000 (20:24 -0700)]
mlx4_core: Avoid double free_icms

On the error path of mlx4_init_hca(), mlx4_close_hca() is called,
followed by mlx4_free_icms() and mlx4_UNMAP_FA().  But both those
functions are also called from mlx4_close_hca(), which leads to a
double free.

Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agomlx4_core: Allocate and map sufficient ICM memory for EQ context
Roland Dreier [Sun, 6 Sep 2009 03:24:49 +0000 (20:24 -0700)]
mlx4_core: Allocate and map sufficient ICM memory for EQ context

The current implementation allocates a single host page for EQ context
memory, which was OK when we only allocated a few EQs.  However, since
we now allocate an EQ for each CPU core, this patch removes the
hard-coded limit (which we exceed with 4 KB pages and 128 byte EQ
context entries with 32 CPUs) and uses the same ICM table code as all
other context tables, which ends up simplifying the code quite a bit
while fixing the problem.

This problem was actually hit in practice on a dual-socket Nehalem box
with 16 real hardware threads and sufficiently odd ACPI tables that it
shows on boot

    SMP: Allowing 32 CPUs, 16 hotplug CPUs

so num_possible_cpus() ends up 32, and mlx4 ends up creating 33 MSI-X
interrupts and 33 EQs.  This mlx4 bug means that mlx4 can't even
initialize at all on this quite mainstream system.

Cc: <stable@kernel.org>
Reported-by: Eli Cohen <eli@mellanox.co.il>
Tested-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mlx4: Annotate CQ locking
Roland Dreier [Sun, 6 Sep 2009 03:24:49 +0000 (20:24 -0700)]
IB/mlx4: Annotate CQ locking

mlx4_ib_lock_cqs()/mlx4_ib_unlock_cqs() are helper functions that
lock/unlock both CQs attached to a QP in the proper order to avoid
AB-BA deadlocks.  Annotate this so sparse can understand what's going
on (and warn us if we misuse these functions).

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agomlx4_core: Remove unnecessary includes of <linux/init.h>
Roland Dreier [Sun, 6 Sep 2009 03:24:49 +0000 (20:24 -0700)]
mlx4_core: Remove unnecessary includes of <linux/init.h>

Lots of mlx4 files with no function annotations included <linux/init.h>
for no reason.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agomlx4_core: Use pci_request_regions()
Roland Dreier [Sun, 6 Sep 2009 03:24:48 +0000 (20:24 -0700)]
mlx4_core: Use pci_request_regions()

The old code used two calls to pci_request_region() to get the two BARs
for the mlx4 device, for no particularly good reason.  Clean up the code
a little by converting this to a single call to pci_request_regions().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/amso1100: Check kmalloc() result in c2_register_device()
Roel Kluin [Sun, 6 Sep 2009 03:24:24 +0000 (20:24 -0700)]
RDMA/amso1100: Check kmalloc() result in c2_register_device()

dev->ibdev.iwcm allocation may fail, prevent a dereference.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/uverbs: Return ENOSYS for unimplemented commands (not EINVAL)
Jack Morgenstein [Sun, 6 Sep 2009 03:24:24 +0000 (20:24 -0700)]
IB/uverbs: Return ENOSYS for unimplemented commands (not EINVAL)

Since the original commit 883a99c7 ("[IB] uverbs: Add a mask of device
methods allowed for userspace"), the uverbs core returns EINVAL for
commands not implemented by a specific low-level driver.

This creates a problem that there is no way to tell the difference
between an unimplemented command and an implemented one which is
incorrectly invoked (which also returns EINVAL).

The fix is to have unimplemented commands return ENOSYS.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/core: Fix send multicast group leave retry
Yossi Etigin [Sun, 6 Sep 2009 03:24:24 +0000 (20:24 -0700)]
IB/core: Fix send multicast group leave retry

Until now, retries were only sent when joining a multicast group. This
patch will adds retries when leaving a multicast group as well.

Signed-off-by: Ron Livne <ronli@voltaire.com>
Signed-off-by: Yossi Etigin <yosefe@voltaire.com>
Acked-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB: Use printk_once() for driver versions
Marcin Slusarz [Sun, 6 Sep 2009 03:24:24 +0000 (20:24 -0700)]
IB: Use printk_once() for driver versions

Replace open-coded reimplementations with printk_once().

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/amso1100: Use %pM conversion specifier
Tobias Klauser [Sun, 6 Sep 2009 03:24:23 +0000 (20:24 -0700)]
RDMA/amso1100: Use %pM conversion specifier

Use the %pM conversion specifier to print a MAC address.

Signed-off-by: Tobias Klauser <klto@zhaw.ch>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB: Use DEFINE_SPINLOCK() for static spinlocks
Roland Dreier [Sun, 6 Sep 2009 03:24:23 +0000 (20:24 -0700)]
IB: Use DEFINE_SPINLOCK() for static spinlocks

Rather than just defining static spinlock_t variables and then
initializing them later in init functions, simply define them with
DEFINE_SPINLOCK() and remove the calls to spin_lock_init().  This cleans
up the source a tad and also shrinks the compiled code; eg on x86-64:

add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-40 (-40)
function                                     old     new   delta
ib_uverbs_init                               336     326     -10
ib_mad_init_module                           147     137     -10
ib_sa_init                                   123     103     -20

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/mad: Check hop count field in directed route MAD to avoid array overflow
Roland Dreier [Sun, 6 Sep 2009 03:24:10 +0000 (20:24 -0700)]
IB/mad: Check hop count field in directed route MAD to avoid array overflow

The hop count field in a directed route MAD is only allowed to be in the
range 0 to 63 (by spec).  Check that this really is the case to avoid
accessing outside the bounds of the hop array.

Reported-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIPoIB: Check multicast address format
Jason Gunthorpe [Sun, 6 Sep 2009 03:23:40 +0000 (20:23 -0700)]
IPoIB: Check multicast address format

Check that the format of multicast link addresses is correct before
taking them from dev->mc_list to priv->multicast_list.  This way we
never try to send a bogus address to the SA, which prevents badness
from erronous 'ip maddr addr add', broken bonding drivers, etc.

Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIPoIB: Drop priv->lock before calling ipoib_send()
Roland Dreier [Sun, 6 Sep 2009 03:23:40 +0000 (20:23 -0700)]
IPoIB: Drop priv->lock before calling ipoib_send()

IPoIB currently must use irqsave locking for priv->lock, since it is
taken from interrupt context in one path.  However, ipoib_send() does
skb_orphan(), and the network stack locking is not IRQ-safe.
Therefore we need to make sure we don't hold priv->lock when calling
ipoib_send() to avoid lockdep warnings (the code was almost certainly
safe in practice, since the only code path that takes priv->lock from
interrupt context would never call into the network stack).

Addresses: http://bugzilla.kernel.org/show_bug.cgi?id=13757
Reported-by: Bart Van Assche <bart.vanassche@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIPoIB: Remove unused <rdma/ib_cache.h> includes
Roland Dreier [Sun, 6 Sep 2009 03:23:38 +0000 (20:23 -0700)]
IPoIB: Remove unused <rdma/ib_cache.h> includes

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/ipath: strncpy() doesn't always NUL-terminate
Roel Kluin [Sun, 6 Sep 2009 03:23:21 +0000 (20:23 -0700)]
IB/ipath: strncpy() doesn't always NUL-terminate

strlcpy() will always null terminate the string.  node_desc is not
guaranteed to be NUL-terminated so just use memcpy().

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/ehca: Fix CQE flags reporting
Joachim Fenkes [Sun, 6 Sep 2009 03:22:55 +0000 (20:22 -0700)]
IB/ehca: Fix CQE flags reporting

The driver was reporting CQE flags in the wrong bit positions, causing
consumers to miss incoming immediate data.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/ehca: Construct MAD redirect replies from request MAD
Joachim Fenkes [Sun, 6 Sep 2009 03:22:55 +0000 (20:22 -0700)]
IB/ehca: Construct MAD redirect replies from request MAD

The old code used a lot of hard-coded values, which might not be valid
in all environments (especially routed fabrics or partitioned
subnets).  Copy as much information as possible from the incoming
request to correct that.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoIB/ehca: Make port autodetect mode the default
Alexander Schmidt [Sun, 6 Sep 2009 03:22:54 +0000 (20:22 -0700)]
IB/ehca: Make port autodetect mode the default

Make port autodetect mode the default for the ehca driver. The
autodetect code has been in the kernel for several releases now and
has proved to be stable.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: Wake up any waiters on peer close/abort
Steve Wise [Sun, 6 Sep 2009 03:22:38 +0000 (20:22 -0700)]
RDMA/cxgb3: Wake up any waiters on peer close/abort

A close/abort while waiting for a wr_ack during connection migration
can cause a hung process in iwch_accept_cr/iwch_reject_cr.

The fix is to set rpl_error/rpl_done and wake up the waiters when we
get a close/abort while in MPA_REQ_RCVD state.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: Don't free endpoints early
Steve Wise [Sun, 6 Sep 2009 03:22:38 +0000 (20:22 -0700)]
RDMA/cxgb3: Don't free endpoints early

- Keep ref on connection request endpoints until either accepted or
  rejected so it doesn't get freed early.

- Endpoint flags now need to be set via atomic bitops because they can
  be set on both the iw_cxgb3 workqueue thread and user disconnect
  threads.

- Don't move out of CLOSING too early due to multiple calls to
  iwch_ep_disconnect.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: Handle port events properly
Steve Wise [Sun, 6 Sep 2009 03:22:38 +0000 (20:22 -0700)]
RDMA/cxgb3: Handle port events properly

Massage the err_handler upcall into an event handler upcall, pass
netdev port events to the cxgb3 ULPs and generate RDMA port events
based on LLD port events.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: Set the appropriate IO channel in rdma_init work requests
Steve Wise [Sun, 6 Sep 2009 03:22:37 +0000 (20:22 -0700)]
RDMA/cxgb3: Set the appropriate IO channel in rdma_init work requests

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoRDMA/cxgb3: iwch_unregister_device leaks memory
Steve Wise [Sun, 6 Sep 2009 03:22:36 +0000 (20:22 -0700)]
RDMA/cxgb3: iwch_unregister_device leaks memory

The iwcm struct mem is never freed.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
12 years agoLinux 2.6.31-rc9 v2.6.31-rc9
Linus Torvalds [Sat, 5 Sep 2009 23:38:12 +0000 (16:38 -0700)]
Linux 2.6.31-rc9

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394...
Linus Torvalds [Sat, 5 Sep 2009 21:59:00 +0000 (14:59 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ieee1394/linux1394-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394-2.6:
  firewire: sbp2: fix freeing of unallocated memory
  firewire: ohci: fix Ricoh R5C832, video reception
  firewire: ohci: fix Agere FW643 and multiple cameras
  firewire: core: fix crash in iso resource management

12 years agopowerpc: Fix i8259 interrupt driver kernel crash on ML510
Roderick Colenbrander [Thu, 3 Sep 2009 15:57:23 +0000 (09:57 -0600)]
powerpc: Fix i8259 interrupt driver kernel crash on ML510

This patch fixes a null pointer exception caused by removal of
'ack()' for level interrupts in the Xilinx interrupt driver.  A recent
change to the xilinx interrupt controller removed the ack hook for
level irqs.

Signed-off-by: Roderick Colenbrander <thunderbird2k@gmail.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMerge git://git.infradead.org/~dwmw2/mtd-2.6.31
Linus Torvalds [Sat, 5 Sep 2009 21:57:04 +0000 (14:57 -0700)]
Merge git://git.infradead.org/~dwmw2/mtd-2.6.31

* git://git.infradead.org/~dwmw2/mtd-2.6.31:
  JFFS2: add missing verify buffer allocation/deallocation
  mtd: nftl: fix offset alignments
  mtd: nftl: write support is broken
  mtd: m25p80: fix null pointer dereference bug

12 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
Linus Torvalds [Sat, 5 Sep 2009 21:56:13 +0000 (14:56 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  block: Allow changing max_sectors_kb above the default 512

12 years agoMerge branch 'fix/oxygen' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Sat, 5 Sep 2009 21:55:30 +0000 (14:55 -0700)]
Merge branch 'fix/oxygen' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'fix/oxygen' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  sound: oxygen: handle cards with missing EEPROM
  sound: oxygen: fix MCLK rate for 192 kHz playback

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Sat, 5 Sep 2009 21:52:41 +0000 (14:52 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  tc: Fix unitialized kernel memory leak
  pkt_sched: Revert tasklet_hrtimer changes.
  net: sk_free() should be allowed right after sk_alloc()
  gianfar: gfar_remove needs to call unregister_netdev()
  ipw2200: firmware DMA loading rework