]> pilppa.com Git - linux-2.6-omap-h63xx.git/log
linux-2.6-omap-h63xx.git
19 years ago[PATCH] namespace.c: fix expiring of detached mount
Miklos Szeredi [Fri, 8 Jul 2005 00:57:26 +0000 (17:57 -0700)]
[PATCH] namespace.c: fix expiring of detached mount

This patch fixes a bug noticed by Al Viro:

   However, we still have a problem here - just what would
   happen if vfsmount is detached while we were grabbing namespace
   semaphore?  Refcount alone is not useful here - we might be held by
   whoever had detached the vfsmount.  IOW, we should check that it's
   still attached (i.e. that mnt->mnt_parent != mnt).  If it's not -
   just leave it alone, do mntput() and let whoever holds it deal with
   the sucker.  No need to put it back on lists.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Cc: <viro@parcelfarce.linux.theplanet.co.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namespace.c: split mark_mounts_for_expiry()
Miklos Szeredi [Fri, 8 Jul 2005 00:57:25 +0000 (17:57 -0700)]
[PATCH] namespace.c: split mark_mounts_for_expiry()

This patch splits the mark_mounts_for_expiry() function.  It's too complex and
too deeply nested, even without the bugfix in the following patch.

Otherwise code is completely the same.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Cc: <viro@parcelfarce.linux.theplanet.co.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namespace.c: cleanup in mark_mounts_for_expiry()
Miklos Szeredi [Fri, 8 Jul 2005 00:57:24 +0000 (17:57 -0700)]
[PATCH] namespace.c: cleanup in mark_mounts_for_expiry()

This patch simplifies mark_mounts_for_expiry() by using detach_mnt() instead
of duplicating everything it does.

It should be an equivalent transformation except for righting the dput/mntput
order.

Al Viro said: "Looks sane".

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Cc: <viro@parcelfarce.linux.theplanet.co.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namespace.c: fix race in mark_mounts_for_expiry()
Miklos Szeredi [Fri, 8 Jul 2005 00:57:24 +0000 (17:57 -0700)]
[PATCH] namespace.c: fix race in mark_mounts_for_expiry()

This patch fixes a race found by Ram in mark_mounts_for_expiry() in
fs/namespace.c.

The bug can only be triggered with simultaneous exiting of a process having
a private namespace, and expiry of a mount from within that namespace.
It's practically impossible to trigger, and I haven't even tried.  But
still, a bug is a bug.

The race happens when put_namespace() is called by another task, while
mark_mounts_for_expiry() is between atomic_read() and get_namespace().  In
that case get_namespace() will be called on an already dead namespace with
unforeseeable results.

The solution was suggested by Al Viro, with his own words:

      Instead of screwing with atomic_read() in there, why don't we
      simply do the following:
       a) atomic_dec_and_lock() in put_namespace()
       b) __put_namespace() called without dropping lock
       c) the first thing done by __put_namespace would be
      struct vfsmount *root = namespace->root;
      namespace->root = NULL;
      spin_unlock(...);
      ....
      umount_tree(root);
      ...
       d) check in mark_... would be simply namespace && namespace->root.

      And we are all set; no screwing around with atomic_read(), no magic
      at all.  Dying namespace gets NULL ->root.
      All changes of ->root happen under spinlock.
      If under a spinlock we see non-NULL ->mnt_namespace, it won't be
      freed until we drop the lock (we will set ->mnt_namespace to NULL
      under that lock before we get to freeing namespace).
      If under a spinlock we see non-NULL ->mnt_namespace and
      ->mnt_namespace->root, we can grab a reference to namespace and be
      sure that it won't go away.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Acked-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namespace.c: fix mnt_namespace clearing
Miklos Szeredi [Fri, 8 Jul 2005 00:57:22 +0000 (17:57 -0700)]
[PATCH] namespace.c: fix mnt_namespace clearing

This patch clears mnt_namespace on unmount.

Not clearing mnt_namespace has two effects:

   1) It is possible to attach a new mount to a detached mount,
      because check_mnt() returns true.

      This means, that when no other references to the detached mount
      remain, it still can't be freed.  This causes a resource leak,
      and possibly un-removable modules.

   2) If mnt_namespace is dereferenced (only in mark_mounts_for_expiry())
      after the namspace has been freed, it can cause an Oops, memory
      corruption, etc.

1) has been tested before and after the patch, 2) is only speculation.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add documentation file
Roland Dreier [Fri, 8 Jul 2005 00:57:21 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add documentation file

Add documentation for InfiniBand userspace verbs.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca user QP support
Roland Dreier [Fri, 8 Jul 2005 00:57:20 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca user QP support

Add support for userspace queue pairs (QPs) to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca user CQ support
Roland Dreier [Fri, 8 Jul 2005 00:57:19 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca user CQ support

Add support for userspace completion queues (CQs) to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca user MR support
Roland Dreier [Fri, 8 Jul 2005 00:57:19 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca user MR support

Add support for userspace memory regions (MRs) to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca user PD support
Roland Dreier [Fri, 8 Jul 2005 00:57:18 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca user PD support

Add support for userspace protection domains (PDs) to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca mmap support
Roland Dreier [Fri, 8 Jul 2005 00:57:17 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca mmap support

Add support for mmap() method to mthca, so that userspace can get access to
doorbell registers.  This allows userspace to get direct access to the HCA for
data path operations.

Each userspace context gets its own copy of the doorbell registers and is only
allowed to use resources that the kernel has given it access to.  In other
words, this is safe.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca user context support
Roland Dreier [Fri, 8 Jul 2005 00:57:16 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca user context support

Add support for managing userspace contexts to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca user doorbell record support
Roland Dreier [Fri, 8 Jul 2005 00:57:16 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca user doorbell record support

Add support for userspace doorbell records to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add mthca ABI header
Roland Dreier [Fri, 8 Jul 2005 00:57:15 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add mthca ABI header

Add the mthca_user.h header file, which defines the device-specific ABI used
by the mthca low-level driver for kernel/user communication.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: hook up Kconfig/Makefile
Roland Dreier [Fri, 8 Jul 2005 00:57:14 +0000 (17:57 -0700)]
[PATCH] IB uverbs: hook up Kconfig/Makefile

Hook up InfiniBand userspace verbs to Kconfig and the make system.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: memory pinning implementation
Roland Dreier [Fri, 8 Jul 2005 00:57:14 +0000 (17:57 -0700)]
[PATCH] IB uverbs: memory pinning implementation

Add support for pinning userspace memory regions and returning a list of pages
in the region.  This includes tracking pinned memory against vm_locked and
preventing unprivileged users from exceeding RLIMIT_MEMLOCK.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: core implementation
Roland Dreier [Fri, 8 Jul 2005 00:57:13 +0000 (17:57 -0700)]
[PATCH] IB uverbs: core implementation

Add the core of the InfiniBand userspace verbs implementation, including
creating character device nodes, dispatching requests from userspace, and
passing event notifications back up to userspace.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: add user verbs ABI header
Roland Dreier [Fri, 8 Jul 2005 00:57:12 +0000 (17:57 -0700)]
[PATCH] IB uverbs: add user verbs ABI header

Add the ib_user_verbs.h header file, which defines the ABI used by InfiniBand
userspace verbs for kernel/user communication.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: update mthca for new API
Roland Dreier [Fri, 8 Jul 2005 00:57:11 +0000 (17:57 -0700)]
[PATCH] IB uverbs: update mthca for new API

Update mthca to compile against the updated API for low-level drivers.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: update kernel midlayer for new API
Roland Dreier [Fri, 8 Jul 2005 00:57:11 +0000 (17:57 -0700)]
[PATCH] IB uverbs: update kernel midlayer for new API

Update kernel InfiniBand midlayer to compile against the updated API for
low-level drivers.  This just amounts to passing NULL for all
userspace-related parameters, and setting userspace-related structure members
to NULL.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB uverbs: core API extensions
Roland Dreier [Fri, 8 Jul 2005 00:57:10 +0000 (17:57 -0700)]
[PATCH] IB uverbs: core API extensions

First of a series of patches which add support for direct userspace access to
InfiniBand hardware -- so-called "userspace verbs." I believe these patches
are ready to merge, but a final review would be useful.

These patches should incorporate all of the feedback from the discussion when
I posted an earlier version back in April (see
http://lkml.org/lkml/2005/4/4/267 for the start of the thread).  In
particular, memory pinned for use by userspace is accounted for in
current->mm->vm_locked and requests to pin memory are checked against
RLIMIT_MEMLOCK.

This patch:

Modify the ib_verbs.h header file with changes required for InfiniBand
userspace verbs support.  We add a few structures to keep track of userspace
context, and extend the driver API so that low-level drivers know when they're
creating resources that will be used from userspace.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] alpha(): pgprot_noncached
Andrew Morton [Fri, 8 Jul 2005 00:57:09 +0000 (17:57 -0700)]
[PATCH] alpha(): pgprot_noncached

The infiniband code expects that the arch implements pgprot_noncached().

We're mapping PCI areas anyway, so this probabyl wasn't needed and we should
make infiniband stop doing that..

Cc: Roland Dreier <rolandd@cisco.com>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] coverity: sunrpc/xprt task null check
KAMBAROV, ZAUR [Fri, 8 Jul 2005 00:57:07 +0000 (17:57 -0700)]
[PATCH] coverity: sunrpc/xprt task null check

In __xprt_lock_write() we check to see if `task' is NULL, but in other places
we just go and dereference it.

`task' shouldn't be NULL anyway, so remove this test.

This defect was found automatically by Coverity Prevent, a static analysis
tool.

Signed-off-by: Zaur Kambarov <zkambarov@coverity.com>
Acked-by: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] coverity: fs/locks.c flp null check
KAMBAROV, ZAUR [Fri, 8 Jul 2005 00:57:06 +0000 (17:57 -0700)]
[PATCH] coverity: fs/locks.c flp null check

We're dereferencing `flp' and then we're testing it for NULLness.

Either the compiler accidentally saved us or the existing null-pointer checdk
is redundant.

This defect was found automatically by Coverity Prevent, a static analysis tool.

Signed-off-by: Zaur Kambarov <zkambarov@coverity.com>
Cc: Matthew Wilcox <willy@debian.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] coverity: fix fbsysfs null pointer check
KAMBAROV, ZAUR [Fri, 8 Jul 2005 00:57:05 +0000 (17:57 -0700)]
[PATCH] coverity: fix fbsysfs null pointer check

Correctly test for a null pointer before going and dereferencing it.

This defect was found automatically by Coverity Prevent, a static analysis
tool.

Signed-off-by: Zaur Kambarov <zkambarov@coverity.com>
Cc: <linux-fbdev-devel@lists.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] cond_resched(): fix bogus might_sleep() warning
Ingo Molnar [Fri, 8 Jul 2005 00:57:04 +0000 (17:57 -0700)]
[PATCH] cond_resched(): fix bogus might_sleep() warning

The BKS might be reacquired before we have dropped PREEMPT_ACTIVE, which
could trigger a second could trigger a second cond_resched() call.  Bug
found by Hirofumi Ogawa.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Keys: Base keyring size on key pointer not key struct
David Howells [Fri, 8 Jul 2005 00:57:03 +0000 (17:57 -0700)]
[PATCH] Keys: Base keyring size on key pointer not key struct

The attached patch makes the keyring functions calculate the new size of a
keyring's payload based on the size of pointer to the key struct, not the size
of the key struct itself.

Signed-Off-By: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] autofs4: mistake in debug print
Ian Kent [Fri, 8 Jul 2005 00:57:02 +0000 (17:57 -0700)]
[PATCH] autofs4: mistake in debug print

Fix debugging printk.

Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] put_compat_shminfo() warning fix
Jesse Millan [Fri, 8 Jul 2005 00:57:01 +0000 (17:57 -0700)]
[PATCH] put_compat_shminfo() warning fix

GCC 4 complains because the function put_compat_shminfo() can't get to its
return statement if there is no error...  If the function does not return
-EFAULT, it doesn't return anything at all.  Looks like a typo.

Signed-off-by: Jesse Millan <jessem@cs.pdx.edu>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ext3 xattr: Don't write to the in-inode xattr space of reserved inodes
Andreas Gruenbacher [Fri, 8 Jul 2005 00:57:00 +0000 (17:57 -0700)]
[PATCH] ext3 xattr: Don't write to the in-inode xattr space of reserved inodes

We are not using the in-inode space for xattrs in reserved inodes because
mkfs.ext3 doesn't initialize it properly.  For those inodes, we set
i_extra_isize to 0.  Make sure that we also don't overwrite the
i_extra_isize field when writing out the inode in that case.  This is for
cleanliness only, and doesn't fix an actual bug.

Signed-off-by: Andreas Gruenbacher <agruen@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mostly_read data section
Christoph Lameter [Fri, 8 Jul 2005 00:56:59 +0000 (17:56 -0700)]
[PATCH] mostly_read data section

Add a new section called ".data.read_mostly" for data items that are read
frequently and rarely written to like cpumaps etc.

If these maps are placed in the .data section then these frequenly read
items may end up in cachelines with data is is frequently updated.  In that
case all processors in an SMP system must needlessly reload the cachelines
again and again containing elements of those frequently used variables.

The ability to share these cachelines will allow each cpu in an SMP system
to keep local copies of those shared cachelines thereby optimizing
performance.

Signed-off-by: Alok N Kataria <alokk@calsoftinc.com>
Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com>
Signed-off-by: Christoph Lameter <christoph@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] propagate __nocast annotations
Alexey Dobriyan [Fri, 8 Jul 2005 00:56:58 +0000 (17:56 -0700)]
[PATCH] propagate __nocast annotations

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] acl kconfig cleanup
Andreas Gruenbacher [Fri, 8 Jul 2005 00:56:57 +0000 (17:56 -0700)]
[PATCH] acl kconfig cleanup

Original patch from Matt Mackall <mpm@selenic.com>

Signed-off-by: Andreas Gruenbacher <agruen@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] page_uptodate locking scalability
Nick Piggin [Fri, 8 Jul 2005 00:56:56 +0000 (17:56 -0700)]
[PATCH] page_uptodate locking scalability

Use a bit spin lock in the first buffer of the page to synchronise asynch
IO buffer completions, instead of the global page_uptodate_lock, which is
showing some scalabilty problems.

Signed-off-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] tty output lossage fix
Roman Zippel [Fri, 8 Jul 2005 00:56:55 +0000 (17:56 -0700)]
[PATCH] tty output lossage fix

The patch fixes a few corner cases around tty line editing with
very long input lines:

- n_tty_receive_char(): don't simply drop eol characters,
  otherwise canon_data isn't increased and the reader isn't woken
  up.

- n_tty_receive_room(): If there is no newline pending and the
  edit buffer is full, allow only a single character to be written
  (until eol is found and the line is flushed), so characters from
  the next line aren't dropped.

- write_chan(): if an incomplete line was written, continue
  writing until write() returns 0, otherwise it might not write
  the eol character to flush the line and the writer goes to sleep
  without ever being woken up.

BTW the core problem is that part of this should be handled in the
receive_buf path, but for this it has to return the number of
written characters, as the amount of written characters may not be
the same as the amount of characters going into the write buffer,
so the receive_room() usage in pty_write() is not really reliable.

Alan said:

The problem looks valid. The behaviour of 'traditional unix' appears to
be the following

If you exceed the line limit then beep and drop the character
Always allow EOL to complete a canonical line input
Always do signal/control processing if enabled

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] xtensa: remove old syscalls
Arnd Bergmann [Fri, 8 Jul 2005 00:56:53 +0000 (17:56 -0700)]
[PATCH] xtensa: remove old syscalls

xtensa is now in -rc1, with the obsolete syscalls still in there, so I
guess this about the last chance to correct the ABI.  Applying the patch
obviously breaks all sorts of user space binaries and probably also
requires the appropriate changes to be made to libc.

On the other hand, if a decision is made to keep the broken interface, it
should at least be a conscious one instead of an oversight.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Chris Zankel <chris@zankel.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] uml: remove winch sem
Paolo 'Blaisorblade' Giarrusso [Fri, 8 Jul 2005 00:56:52 +0000 (17:56 -0700)]
[PATCH] uml: remove winch sem

Replace a semaphore (winch_handler_sem) used in atomic code with a
spinlock, and reduces as needed the amount of protected code to the bare
minimum (for instance no kmalloc calls are needed).

This fixes the last problems with spinlocking (in UP mode with DEBUG
options); the semaphore, taken inside spinlocks, caused a "spin_lock was
already locked" warning, without this patch.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] uml: restore hppfs support
Paolo 'Blaisorblade' Giarrusso [Fri, 8 Jul 2005 00:56:51 +0000 (17:56 -0700)]
[PATCH] uml: restore hppfs support

Some time ago a trivial patch broke HPPFS (one var became a pointer, not
all uses were updated).  It wasn't fixed at that time because not very
used, now it's been requested so I've fixed this, and it has been tested
positively (at least partially).

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] uml: Proper clone support for skas0
Bodo Stroesser [Fri, 8 Jul 2005 00:56:50 +0000 (17:56 -0700)]
[PATCH] uml: Proper clone support for skas0

This patch implements the clone-stub mechanism, which allows skas0 to run
with proc_mm==0, even if the clib in UML uses modify_ldt.

Note: There is a bug in skas3.v7 host patch, that avoids UML-skas from
running properly on a SMP-box.  In full skas3, I never really saw problems,
but in skas0 they showed up.

More commentary by jdike - What this patch does is makes sure that the host
parent of each new host process matches the UML parent of the corresponding
UML process.  This ensures that any changed LDTs are inherited.  This is
done by having clone actually called by the UML process from its stub,
rather than by the kernel.  We have special syscall stubs that are loaded
onto the stub code page because that code must be completely
self-contained.  These stubs are given C interfaces, and used like normal C
functions, but there are subtleties.  Principally, we have to be careful
about stack variables in stub_clone_handler after the clone.  The code is
written so that there aren't any - everything boils down to a fixed
address.  If there were any locals, references to them after the clone
would be wrong because the stack just changed.

Signed-off-by: Bodo Stroesser <bstroesser@fujitsu-siemens.com>
Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] uml: skas0 - separate kernel address space on stock hosts
Jeff Dike [Fri, 8 Jul 2005 00:56:49 +0000 (17:56 -0700)]
[PATCH] uml: skas0 - separate kernel address space on stock hosts

UML has had two modes of operation - an insecure, slow mode (tt mode) in
which the kernel is mapped into every process address space which requires
no host kernel modifications, and a secure, faster mode (skas mode) in
which the UML kernel is in a separate host address space, which requires a
patch to the host kernel.

This patch implements something very close to skas mode for hosts which
don't support skas - I'm calling this skas0.  It provides the security of
the skas host patch, and some of the performance gains.

The two main things that are provided by the skas patch, /proc/mm and
PTRACE_FAULTINFO, are implemented in a way that require no host patch.

For the remote address space changing stuff (mmap, munmap, and mprotect),
we set aside two pages in the process above its stack, one of which
contains a little bit of code which can call mmap et al.

To update the address space, the system call information (system call
number and arguments) are written to the stub page above the code.  The
%esp is set to the beginning of the data, the %eip is set the the start of
the stub, and it repeatedly pops the information into its registers and
makes the system call until it sees a system call number of zero.  This is
to amortize the cost of the context switch across multiple address space
updates.

When the updates are done, it SIGSTOPs itself, and the kernel process
continues what it was doing.

For a PTRACE_FAULTINFO replacement, we set up a SIGSEGV handler in the
child, and let it handle segfaults rather than nullifying them.  The
handler is in the same page as the mmap stub.  The second page is used as
the stack.  The handler reads cr2 and err from the sigcontext, sticks them
at the base of the stack in a faultinfo struct, and SIGSTOPs itself.  The
kernel then reads the faultinfo and handles the fault.

A complication on x86_64 is that this involves resetting the registers to
the segfault values when the process is inside the kill system call.  This
breaks on x86_64 because %rcx will contain %rip because you tell SYSRET
where to return to by putting the value in %rcx.  So, this corrupts $rcx on
return from the segfault.  To work around this, I added an
arch_finish_segv, which on x86 does nothing, but which on x86_64 ptraces
the child back through the sigreturn.  This causes %rcx to be restored by
sigreturn and avoids the corruption.  Ultimately, I think I will replace
this with the trick of having it send itself a blocked signal which will be
unblocked by the sigreturn.  This will allow it to be stopped just after
the sigreturn, and PTRACE_SYSCALLed without all the back-and-forth of
PTRACE_SYSCALLing it through sigreturn.

This runs on a stock host, so theoretically (and hopefully), tt mode isn't
needed any more.  We need to make sure that this is better in every way
than tt mode, though.  I'm concerned about the speed of address space
updates and page fault handling, since they involve extra round-trips to
the child.  We can amortize the round-trip cost for large address space
updates by writing all of the operations to the data page and having the
child execute them all at the same time.  This will help fork and exec, but
not page faults, since they involve only one page.

I can't think of any way to help page faults, except to add something like
PTRACE_FAULTINFO to the host.  There is PTRACE_SIGINFO, but UML doesn't use
siginfo for SIGSEGV (or anything else) because there isn't enough
information in the siginfo struct to handle page faults (the faulting
operation type is missing).  Adding that would make PTRACE_SIGINFO a usable
equivalent to PTRACE_FAULTINFO.

As for the code itself:

- The system call stub is in arch/um/kernel/sys-$(SUBARCH)/stub.S.  It is
  put in its own section of the binary along with stub_segv_handler in
  arch/um/kernel/skas/process.c.  This is manipulated with run_syscall_stub
  in arch/um/kernel/skas/mem_user.c.  syscall_stub will execute any system
  call at all, but it's only used for mmap, munmap, and mprotect.

- The x86_64 stub calls sigreturn by hand rather than allowing the normal
  sigreturn to happen, because the normal sigreturn is a SA_RESTORER in
  UML's address space provided by libc.  Needless to say, this is not
  available in the child's address space.  Also, it does a couple of odd
  pops before that which restore the stack to the state it was in at the
  time the signal handler was called.

- There is a new field in the arch mmu_context, which is now a union.
  This is the pid to be manipulated rather than the /proc/mm file
  descriptor.  Code which deals with this now checks proc_mm to see whether
  it should use the usual skas code or the new code.

- userspace_tramp is now used to create a new host process for every UML
  process, rather than one per UML processor.  It checks proc_mm and
  ptrace_faultinfo to decide whether to map in the pages above its stack.

- start_userspace now makes CLONE_VM conditional on proc_mm since we need
  separate address spaces now.

- switch_mm_skas now just sets userspace_pid[0] to the new pid rather
  than PTRACE_SWITCH_MM.  There is an addition to userspace which updates
  its idea of the pid being manipulated each time around the loop.  This is
  important on exec, when the pid will change underneath userspace().

- The stub page has a pte, but it can't be mapped in using tlb_flush
  because it is part of tlb_flush.  This is why it's required for it to be
  mapped in by userspace_tramp.

Other random things:

- The stub section in uml.lds.S is page aligned.  This page is written
  out to the backing vm file in setup_physmem because it is mapped from
  there into user processes.

- There's some confusion with TASK_SIZE now that there are a couple of
  extra pages that the process can't use.  TASK_SIZE is considered by the
  elf code to be the usable process memory, which is reasonable, so it is
  decreased by two pages.  This confuses the definition of
  USER_PGDS_IN_LAST_PML4, making it too small because of the rounding down
  of the uneven division.  So we round it to the nearest PGDIR_SIZE rather
  than the lower one.

- I added a missing PT_SYSCALL_ARG6_OFFSET macro.

- um_mmu.h was made into a userspace-usable file.

- proc_mm and ptrace_faultinfo are globals which say whether the host
  supports these features.

- There is a bad interaction between the mm.nr_ptes check at the end of
  exit_mmap, stack randomization, and skas0.  exit_mmap will stop freeing
  pages at the PGDIR_SIZE boundary after the last vma.  If the stack isn't
  on the last page table page, the last pte page won't be freed, as it
  should be since the stub ptes are there, and exit_mmap will BUG because
  there is an unfreed page.  To get around this, TASK_SIZE is set to the
  next lowest PGDIR_SIZE boundary and mm->nr_ptes is decremented after the
  calls to init_stub_pte.  This ensures that we know the process stack (and
  all other process mappings) will be below the top page table page, and
  thus we know that mm->nr_ptes will be one too many, and can be
  decremented.

Things that need fixing:

- We may need better assurrences that the stub code is PIC.

- The stub pte is set up in init_new_context_skas.

- alloc_pgdir is probably the right place.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] pm: clean up process.c
Pavel Machek [Fri, 8 Jul 2005 00:56:45 +0000 (17:56 -0700)]
[PATCH] pm: clean up process.c

freezeable() already tests for TRACED/STOPPED processes, no need to do it
twice.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] swsusp: fix error handling
Pavel Machek [Fri, 8 Jul 2005 00:56:44 +0000 (17:56 -0700)]
[PATCH] swsusp: fix error handling

Fix error handling and whitespace in swsusp.c.  swsusp_free() was called when
there was nothing allocating, leading to oops.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] pm: Fix resume from initrd
Pavel Machek [Fri, 8 Jul 2005 00:56:43 +0000 (17:56 -0700)]
[PATCH] pm: Fix resume from initrd

Move device name resolution code around so that it is not called from
resume-from-initrd.  name_to_dev_t may be unavailable at that point.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] pm: fix u32 vs. pm_message_t confusion in cpufreq
Bernard Blackham [Fri, 8 Jul 2005 00:56:42 +0000 (17:56 -0700)]
[PATCH] pm: fix u32 vs. pm_message_t confusion in cpufreq

Fix u32 vs pm_message_t confusion in cpufreq.

Signed-off-by: Bernard Blackham <bernard@blackham.com.au>
Signed-off-by: Pavel Machek <pavel@suse.cz>
Cc: Dave Jones <davej@codemonkey.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] pm: more u32 vs. pm_message_t fixes
Pavel Machek [Fri, 8 Jul 2005 00:56:40 +0000 (17:56 -0700)]
[PATCH] pm: more u32 vs. pm_message_t fixes

Few more u32 vs. pm_message_t fixes.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Fix up non-NUMA breakage in mmzone.h
Dave Jones [Fri, 8 Jul 2005 00:56:39 +0000 (17:56 -0700)]
[PATCH] Fix up non-NUMA breakage in mmzone.h

If CONFIG_NUMA isn't set, we use the define in <linux/mmzone.h> for
early_pfn_to_nid (which defines it to 0).

Because of this, the prototype needs to move inside the CONFIG_NUMA too, or
anal gcc's get really confused.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Clean up numa defines in mmzone.h
Dave Jones [Fri, 8 Jul 2005 00:56:39 +0000 (17:56 -0700)]
[PATCH] Clean up numa defines in mmzone.h

The recent cleanups to asm-i386/mmzone.h were suboptimal nesting an ifdef of
the same symbol.  This patch removes some of the ifdef'ery to make things more
readable again.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] MTRR suspend/resume cleanup
Shaohua Li [Fri, 8 Jul 2005 00:56:38 +0000 (17:56 -0700)]
[PATCH] MTRR suspend/resume cleanup

There has been some discuss about solving the SMP MTRR suspend/resume
breakage, but I didn't find a patch for it.  This is an intent for it.  The
basic idea is moving mtrr initializing into cpu_identify for all APs (so it
works for cpu hotplug).  For BP, restore_processor_state is responsible for
restoring MTRR.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FRV: Add defconfig
Adrian Bunk [Fri, 8 Jul 2005 00:56:36 +0000 (17:56 -0700)]
[PATCH] FRV: Add defconfig

This patch by Yoshihiro MATSUYAMA (already ACK'ed by David Howells) adds a
defconfig for the frv arch.

Signed-Off-By: Yoshihiro MATSUYAMA <y.matsu@jp.fujitsu.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: silence perfmon exception warnings
Anton Blanchard [Fri, 8 Jul 2005 00:56:36 +0000 (17:56 -0700)]
[PATCH] ppc64: silence perfmon exception warnings

We dont need to use the PERFMON exception on POWER5, in fact the firmware
returns an error.  Due to this just remove the warning.

Also now that we have proper runlatch support we can remove the bootup
hack.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Be consistent about printing which idle loop we're using
Michael Ellerman [Fri, 8 Jul 2005 00:56:35 +0000 (17:56 -0700)]
[PATCH] ppc64: Be consistent about printing which idle loop we're using

Not sure if we really need this, but it was handy to know which iSeries loop I
was testing.

Be consistent about printing which idle loop we're using, with this patch we
cover all cases.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: fix compile warning
Anton Blanchard [Fri, 8 Jul 2005 00:56:34 +0000 (17:56 -0700)]
[PATCH] ppc64: fix compile warning

Fix a compile warning introduced by the previous patches.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: idle fixups
Anton Blanchard [Fri, 8 Jul 2005 00:56:33 +0000 (17:56 -0700)]
[PATCH] ppc64: idle fixups

- remove some unnecessary includes
- add runlatch support
- no need to use raw_smp_processor_id any more, current preempt debug
  logic checks for processes that are bound to one cpu.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: pSeries idle fixups
Anton Blanchard [Fri, 8 Jul 2005 00:56:33 +0000 (17:56 -0700)]
[PATCH] ppc64: pSeries idle fixups

- separate out sleep logic in dedicated_idle, it was so far indented
  that it got squashed against the right side of the screen.
- add runlatch support, looping on runlatch disable.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: iSeries idle fixups
Anton Blanchard [Fri, 8 Jul 2005 00:56:32 +0000 (17:56 -0700)]
[PATCH] ppc64: iSeries idle fixups

- remove min/max yield time, we dont use the values anywhere
- separate shared and dedicated idle loops
- check need_resched again with irqs off to avoid sleeping with pending work
- continually set runlatch off in idle loop, this means we dont need to
  turn the runlatch off on exception exit and suffer that associated
  cost for all exceptions. (A future patch will turn the runlatch on at
  exception entry)

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Remove obsolete idle_setup()
Michael Ellerman [Fri, 8 Jul 2005 00:56:31 +0000 (17:56 -0700)]
[PATCH] ppc64: Remove obsolete idle_setup()

Now that the idle loop is configured by each platform we don't need
idle_setup() anymore.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Fixup platforms for new ppc_md.idle
Michael Ellerman [Fri, 8 Jul 2005 00:56:30 +0000 (17:56 -0700)]
[PATCH] ppc64: Fixup platforms for new ppc_md.idle

This patch fixes up iSeries, pSeries, pmac and maple to set the correct idle
function for each platform.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Move pSeries idle functions into pSeries_setup.c
Michael Ellerman [Fri, 8 Jul 2005 00:56:29 +0000 (17:56 -0700)]
[PATCH] ppc64: Move pSeries idle functions into pSeries_setup.c

dedicated_idle() and shared_idle() are only used by pSeries, so move them into
pSeries_setup.c

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Move iSeries_idle() into iSeries_setup.c
Michael Ellerman [Fri, 8 Jul 2005 00:56:29 +0000 (17:56 -0700)]
[PATCH] ppc64: Move iSeries_idle() into iSeries_setup.c

Move iSeries_idle() into iSeries_setup.c, no one else needs to know about it.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Make idle_loop a ppc_md function
Michael Ellerman [Fri, 8 Jul 2005 00:56:28 +0000 (17:56 -0700)]
[PATCH] ppc64: Make idle_loop a ppc_md function

This patch adds an idle member to the ppc_md structure and calls it from
cpu_idle().  If a platform leaves ppc_md.idle as null it will get the default
idle loop default_idle().

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Use hvc_get_chars in hvsi code
Milton Miller [Fri, 8 Jul 2005 00:56:27 +0000 (17:56 -0700)]
[PATCH] hvc_console: Use hvc_get_chars in hvsi code

Now that hvc_get_chars doesn't strip NULs, hvsi doesn't have to duplicate it.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Separate the NUL character filtering from get_hvc_chars
Milton Miller [Fri, 8 Jul 2005 00:56:26 +0000 (17:56 -0700)]
[PATCH] hvc_console: Separate the NUL character filtering from get_hvc_chars

Separate the NUL character filtering from get_hvc_chars.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Register ops when setting up hvc_console
Milton Miller [Fri, 8 Jul 2005 00:56:25 +0000 (17:56 -0700)]
[PATCH] hvc_console: Register ops when setting up hvc_console

When registering the hvc console port, register a list of ops (read and write)
to go with it, instead of calling fixed function names.

This allows different ports to encode the data differently.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Separate hvc_console and vio code 2
Milton Miller [Fri, 8 Jul 2005 00:56:24 +0000 (17:56 -0700)]
[PATCH] hvc_console: Separate hvc_console and vio code 2

Remove all the vio device driver code from hvc_console.c

This will allow us to separate hvsi, hvc, and allow hvc_console to be used
without the ppc64 vio layer.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Separate hvc_console and vio code
Milton Miller [Fri, 8 Jul 2005 00:56:24 +0000 (17:56 -0700)]
[PATCH] hvc_console: Separate hvc_console and vio code

Separate the console setup routines of the hvc_console and the vio layer.

Remove the call to find_init_vty from hvc_console.c.

Fail the setup routine if the console doesn't exist, but register the console
again when the specified channel is instantiated.  This scheme maintains the
print buffer semantics while eliminating callout and call back for the console
code.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Add some sanity checks
Milton Miller [Fri, 8 Jul 2005 00:56:23 +0000 (17:56 -0700)]
[PATCH] hvc_console: Add some sanity checks

Check if a vterm was registered before accepting it as a console.
Check that a slot hasn't been probed with a tty in hvc_instantiate().
Check that a slot hasn't been free'ed when handing out console device.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Statically initialize the vtermnos array
Milton Miller [Fri, 8 Jul 2005 00:56:22 +0000 (17:56 -0700)]
[PATCH] hvc_console: Statically initialize the vtermnos array

Statically initialize the vtermnos array.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: remove num_vterms and some dead code
Milton Miller [Fri, 8 Jul 2005 00:56:21 +0000 (17:56 -0700)]
[PATCH] hvc_console: remove num_vterms and some dead code

num_vterms hasn't been used since the hotplug support went in.  Also, remove a
dead code line from a list_for_each_entry conversion.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Add missing include
Milton Miller [Fri, 8 Jul 2005 00:56:21 +0000 (17:56 -0700)]
[PATCH] hvc_console: Add missing include

hvc_console checks MAGIC_SYSRQ and XMON config vars.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Unregister the console in the exit routine.
Milton Miller [Fri, 8 Jul 2005 00:56:20 +0000 (17:56 -0700)]
[PATCH] hvc_console: Unregister the console in the exit routine.

Be thorough in our exit routine, since it says it is there to be so.
Unregistering without registering is safe (checked in 2.6.10).

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: MAGIC_SYSRQ should only be on console channel
Milton Miller [Fri, 8 Jul 2005 00:56:19 +0000 (17:56 -0700)]
[PATCH] hvc_console: MAGIC_SYSRQ should only be on console channel

Guard the MAGIC_SYSRQ ^O to be just on the console channel.  Make the other
channels more transparent.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Dont always kick the poll thread in interrupt
Milton Miller [Fri, 8 Jul 2005 00:56:18 +0000 (17:56 -0700)]
[PATCH] hvc_console: Dont always kick the poll thread in interrupt

Have the hvc console code try to pull characters immediately when receiving an
interrupt, and kick the poll thread only if the immediate poll indicates it
needed a call back to do more work.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Match vio and console devices using vterm numbers
Milton Miller [Fri, 8 Jul 2005 00:56:17 +0000 (17:56 -0700)]
[PATCH] hvc_console: Match vio and console devices using vterm numbers

Use the vterm numbers to match the vio devices being probed with the indices
already allocated via the console initcall function hvc_find_vtys.

The old code required hvc_find_vtys to "guess" the matching devices the vio
subsystem would find and its probe order.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hvc_console: Rearrange code
Milton Miller [Fri, 8 Jul 2005 00:56:16 +0000 (17:56 -0700)]
[PATCH] hvc_console: Rearrange code

Milton Miller has done a lot of work to clean up our hvc_console code.

One of the important things the following patch series does is separate the
VIO layer from the hvc_console code.  With the VIO specific code removed any
ppc64 platform, or even any architecture, can use hvc_console as a generic
polling console.  You simply have to supply a get_chars and put_chars method
and hvc_console does the rest of the work.  You can even use it for an
interrupt driven console.

This patch:

Rearrange the code in drivers/char/hvc_console.c to make future patches
smaller.  No actual code changes, just ordering of the functions in the file.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: remove duplicate syscall reservation
Anton Blanchard [Fri, 8 Jul 2005 00:56:15 +0000 (17:56 -0700)]
[PATCH] ppc64: remove duplicate syscall reservation

We already have a prototype for sys_remap_file_pages (239) so there is no need
to reserve it twice.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: add ioprio syscalls
Anton Blanchard [Fri, 8 Jul 2005 00:56:15 +0000 (17:56 -0700)]
[PATCH] ppc64: add ioprio syscalls

- Clean up sys32_getpriority comment.
- Add ioprio syscalls, and sign extend 32bit versions.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: sys_ppc32.c cleanups
Anton Blanchard [Fri, 8 Jul 2005 00:56:14 +0000 (17:56 -0700)]
[PATCH] ppc64: sys_ppc32.c cleanups

Remove some unnecessary includes, an out of date comment and a prototype for
sys_timer_create (which is now in syscalls.h)

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] move ioprio syscalls into syscalls.h
Anton Blanchard [Fri, 8 Jul 2005 00:56:13 +0000 (17:56 -0700)]
[PATCH] move ioprio syscalls into syscalls.h

- Make ioprio syscalls return long, like set/getpriority syscalls.
- Move function prototypes into syscalls.h so we can pick them up in the
  32/64bit compat code.

Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Turn runlatch on in exception entry
Anton Blanchard [Fri, 8 Jul 2005 00:56:12 +0000 (17:56 -0700)]
[PATCH] ppc64: Turn runlatch on in exception entry

Enable the runlatch at the start of each exception.  Unfortunately we are out
of space in the 0x300 handler, so I added it a bit later.

The SPR write is fairly expensive, perhaps we should cache the runlatch state
in the paca and avoid the write when possible.

We don't need to turn the runlatch off, we do that in the idle loop.  Better
to take the hit in the idle loop than for each exception exit.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Fix runlatch code to work on pseries machines
Anton Blanchard [Fri, 8 Jul 2005 00:56:11 +0000 (17:56 -0700)]
[PATCH] ppc64: Fix runlatch code to work on pseries machines

Not all ppc64 CPUs have the CTRL SPR, so we need a cputable feature for it.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: use c99 initialisers in cputable code
Anton Blanchard [Fri, 8 Jul 2005 00:56:11 +0000 (17:56 -0700)]
[PATCH] ppc64: use c99 initialisers in cputable code

Use c99 initialisers in the cputable code.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: vdso32: fix link errors after recent toolchain changes
Olaf Hering [Fri, 8 Jul 2005 00:56:09 +0000 (17:56 -0700)]
[PATCH] ppc64: vdso32: fix link errors after recent toolchain changes

Patch from <amodra@bigpond.net.au>,
http://sources.redhat.com/bugzilla/show_bug.cgi?id=1042

/usr/bin/ld: arch/ppc64/kernel/vdso32/vdso32.so: The first section in the
PT_DYNAMIC segment is not the .dynamic section

Signed-off-by: Olaf Hering <olh@suse.de>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Add new PHY to sungem
Benjamin Herrenschmidt [Fri, 8 Jul 2005 00:56:09 +0000 (17:56 -0700)]
[PATCH] ppc64: Add new PHY to sungem

This patch adds support for some new PHY models to sungem as used on some
recent Apple iMac G5 models.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] kbuild: build a single module using 'make dir/module.ko'
Sam Ravnborg [Fri, 8 Jul 2005 00:56:08 +0000 (17:56 -0700)]
[PATCH] kbuild: build a single module using 'make dir/module.ko'

Using the syntax:
make dir/module.ko

kbuild now allows one to build a module including the final link stage.
This is usefull when one only wants to compile a single module and thus do
not have to wait until a full kernel has finished compiling.  Tested by:
randy_dunlap <rdunlap@xenotime.net>

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] kbuild: build TAGS problem with O=
George Anzinger [Fri, 8 Jul 2005 00:56:06 +0000 (17:56 -0700)]
[PATCH] kbuild: build TAGS problem with O=

  make O=/dir TAGS

  fails with:

    MAKE   TAGS
  find: security/selinux/include: No such file or directory
  find: include: No such file or directory
  find: include/asm-i386: No such file or directory
  find: include/asm-generic: No such file or directory

  The problem is in this line:
  ifeq ($(KBUILD_OUTPUT),)

KBUILD_OUTPUT is not defined (ever) after make reruns itself.  This line is
used in the TAGS, tags, and cscope makes.

Signed-off-by: George Anzinger <george@mvista.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mm: quieten OOM killer noise
Anton Blanchard [Fri, 8 Jul 2005 00:56:06 +0000 (17:56 -0700)]
[PATCH] mm: quieten OOM killer noise

We now print statistics when invoking the OOM killer, however this
information is not rate limited and you can get into situations where the
console is continually spammed.

For example, when a task is exiting the OOM killer will simply return
(waiting for that task to exit and clear up memory).  If the VM continually
calls back into the OOM killer we get thousands of copies of show_mem() on
the console.

Use printk_ratelimit() to quieten it.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] remove completly bogus comment inside __alloc_pages() try_to_free_pages handling
Marcelo Tosatti [Fri, 8 Jul 2005 00:56:05 +0000 (17:56 -0700)]
[PATCH] remove completly bogus comment inside __alloc_pages() try_to_free_pages handling

Remove completly bogus comment from did_some_progress != 0 handling (that
same comment is a few lines below on did_some_progress = 0 case, where it
belongs).

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] print order information when OOM killing
Marcelo Tosatti [Fri, 8 Jul 2005 00:56:04 +0000 (17:56 -0700)]
[PATCH] print order information when OOM killing

Dump the current allocation order when OOM killing.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] export generic_drop_inode() to modules
Mark Fasheh [Fri, 8 Jul 2005 00:56:03 +0000 (17:56 -0700)]
[PATCH] export generic_drop_inode() to modules

OCFS2 wants to mark an inode which has been orphaned by another node so
that during final iput it takes the correct path through the VFS and can
pass through the OCFS2 delete_inode callback.  Since i_nlink can get out of
date with other nodes, the best way I see to accomplish this is by clearing
i_nlink on those inodes at drop_inode time.  Other than this small amount
of work, nothing different needs to happen, so I think it would be cleanest
to be able to just call generic_drop_inode at the end of the OCFS2
drop_inode callback.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] i2o: config-osm build fix
Andrew Morton [Fri, 8 Jul 2005 00:56:02 +0000 (17:56 -0700)]
[PATCH] i2o: config-osm build fix

Various stuff missing on alpha:

drivers/message/i2o/config-osm.c:35: error: field `fops' has incomplete type
drivers/message/i2o/config-osm.c: In function `sysfs_create_fops_file':
drivers/message/i2o/config-osm.c:71: error: storage size of `tmp' isn't known
drivers/message/i2o/config-osm.c:78: error: dereferencing pointer to incomplete type
drivers/message/i2o/config-osm.c:81: error: dereferencing pointer to incomplete type

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] iounmap debugging
Andrew Morton [Fri, 8 Jul 2005 00:56:02 +0000 (17:56 -0700)]
[PATCH] iounmap debugging

We get sporadic reports of `__iounmap: bad address' coming out.  Add a
dump_stack() to find the culprit.

Try to identify which subsystem is having iounmap() problems.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] uml: kill some useless vmalloc tlb flushing
Jeff Dike [Fri, 8 Jul 2005 00:56:00 +0000 (17:56 -0700)]
[PATCH] uml: kill some useless vmalloc tlb flushing

There is absolutely no reason to flush the kernel's VM area during a
tlb_flush_mm.

This results in a noticable performance increase in the kernel build
benchmark.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: Paolo Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Thu, 7 Jul 2005 17:24:51 +0000 (10:24 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/aegl/linux-2.6

19 years ago[IA64] - Disable tiocx driver on non-SN systems
Jack Steiner [Thu, 7 Jul 2005 16:14:00 +0000 (09:14 -0700)]
[IA64] - Disable tiocx driver on non-SN systems

Disable the tiocx driver on non-SN systems.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
19 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Thu, 7 Jul 2005 05:15:13 +0000 (22:15 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6

19 years ago[IA64] fix generic/up builds
Tony Luck [Thu, 7 Jul 2005 01:18:10 +0000 (18:18 -0700)]
[IA64] fix generic/up builds

Jesse Barnes provided the original version of this patch months ago, but
other changes kept conflicting with it, so it got deferred.  Greg Edwards
dug it out of obscurity just over a week ago, and almost immediately
another conflicting patch appeared (Bob Picco's memory-less nodes).

I've resolved the conflicts and got it running again.  CONFIG_SGI_TIOCX
is set to "y" in defconfig, which causes a Tiger to not boot (oops in
tiocx_init).  But that can be resolved later ... get this in now before it
gets stale again.

Signed-off-by: Tony Luck <tony.luck@intel.com>
19 years agoMerge rsync://rsync.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Thu, 7 Jul 2005 00:04:06 +0000 (17:04 -0700)]
Merge rsync://rsync.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6

19 years ago[PATCH] yet another fix for setup-bus.c/x86 merge
Ivan Kokshaysky [Wed, 6 Jul 2005 23:07:56 +0000 (03:07 +0400)]
[PATCH] yet another fix for setup-bus.c/x86 merge

There is a slight disagreement between setup-bus.c code and traditional
x86 PCI setup wrt which recourses are invalid vs resources that are free
for further allocations.

In particular, in the setup-bus.c, if we failed to allocate some resource,
we nullify "start" and "flags" fields, but *not* the "end" one.

But x86 pcibios_enable_resources() does the following check:

if (!r->start && r->end) {
printk(KERN_ERR "PCI: Device %s not available because of resource collisions\n", pci_name(dev));
return -EINVAL;

which means that the device owning the offending resource cannot be
enabled.

In particular, this breaks cardbus behind the normal decode p2p bridge -
the cardbus code from setup-bus.c requests rather large IO and MEM
windows, and if it fails, the socket is completely unavailable.  Which
is wrong, as the yenta code is capable to allocate smaller windows.

Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[IA64] memory-less-nodes repost
bob.picco [Thu, 30 Jun 2005 16:52:00 +0000 (09:52 -0700)]
[IA64] memory-less-nodes repost

I reworked how nodes with only CPUs are treated.  The patch below seems
simpler to me and has eliminated the complicated routine
reassign_cpu_only_nodes.  There isn't any longer the requirement
to modify ACPI NUMA information which was in large part the
complexity introduced in reassign_cpu_only_nodes.

This patch will produce a different number of nodes. For example,
reassign_cpu_only_nodes would reduce two CPUonly nodes and one memory node
configuration to one memory+CPUs node configuration.  This patch
doesn't change the number of nodes which means the user will see three.  Two
nodes without memory and one node with all the memory.

While doing this patch, I noticed that early_nr_phys_cpus_node isn't serving
any useful purpose.  It is called once in find_pernode_space but the value
isn't used to computer pernode space.

Signed-off-by: bob.picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
19 years ago[IA64] Make ia64 die() preempt safe
<> [Sat, 2 Jul 2005 06:27:00 +0000 (23:27 -0700)]
[IA64] Make ia64 die() preempt safe

Signed-off-by: Keith Owens <kaos@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>