From: Benjamin Herrenschmidt <benh@kernel.crashing.org>

iSeries HyperVisor is doing some evilery when inserting PTEs that I didn't
properly account for when rewriting that code, causing iSeries box to blow
up regulary.

address_spaces have a `dirtied_when' jiffies field which records the time at
which the mapping was first dirtied.  This is used for the periodic writeback
(kupdate) function.  It is also used to prevent livelocks in the writeback
code: don't try to write back objects which were dirtied after
sync_sb_inodes() was called.

It used to be the case that dirtied_when == 0 had magical properties, so
there is code in there which avoids accidentally setting dirtied_when to zero
at jiffy wrap time.  We just set it to jiffies|1.

Unfortunately, jiffies|1 is in the future.  So under some rare timing
circumstances (inode dirtied within one jiffy of umount) the livelock
avoidance code in sync_sb_inodes() can accidentally trigger and we fail to
write an inode out, resulting in filesytem corruption on ext2-style
filesystems.  Normally, nobody dirties a file within a millisecond of umount,
so it was not noticed.

It is no longer the case that (address_space.dirtied_when == 0) has special
meaning, so we can just remove all that code and fix the bug.

This fixes the PowerPC Virtual Ethernet driver that got broken by the
recent ppc64 iommu patch...

into ppc970.osdl.org:/home/torvalds/v2.5/linux

into ppc970.osdl.org:/home/torvalds/v2.5/linux

On Mon, Nov 17, 2003 at 02:57:44PM -0800, Jean Tourrilhes wrote:
> On Sun, Nov 16, 2003 at 11:33:01PM +0000, Russell King wrote:
> > Jean,
> >
> > Could you please arrange for this patch to be tested?  It covers all
> > the PCMCIA wireless drivers in 2.6.0-test9; I haven't been able to
> > test.
>
> 	Dual P500, kernel 2.6.0-test9 + your patch.
> -------------------------------------------------
>...
> 	Few comments :
> 	o Yanked the cards out of the socket, cardmgr did ifconfig
> down and removal of module automatically.
> 	o I did not see any patch for the Orinoco driver, but I did
> not check if it needs one. Did not look at HostAP either...
> 	o wl3501_cs : ask Arnaldo.
> 	o With 2.6.0-test9, I had instances of "can't read CIS". I
> actually had to reboot to perform the test.
> 	o The kobject Oops is something I think I saw before. I never
> remove my Aironet card because I don't want to reboot :-( This time,
> it seems to have survived better, so I see that as progress ;-)

This patch seems to have been forgotten about, so here it is again,
this time against 2.6.4-rc1.  This really needs to be send upstream,
though I don't know where we got to with this one.  Maybe we should
put it in akpm's tree for a while?

It's essentially more of the "don't deadlock if you pull out a
PCMCIA network card and it's currently configured up" patch which
went in recently for PCMCIA NIC cards.

>>>>> "Pawel" == Pawe  <Pawe> writes:

Pawel> On 2.6.3 it just Oopsed and didn't work. On 2.6.4-rc1 it does
Pawel> not oops but it doesn't work (neither as module nor as build in kernel).

Note that hp100 in 2.6.4-rc1 is still broken wrt EISA probing, since
it lacks the terminating entry in the EISA ID list. This one-liner
should take care of random crashes that are often reported to LKML :

===== drivers/net/hp100.c 1.24 vs edited =====

into ppc970.osdl.org:/home/torvalds/v2.5/linux

into ppc970.osdl.org:/home/torvalds/v2.5/linux

To hotplug-remove virtual devices, we need vio_find_node() so we have a
pointer to pass to vio_unregister_device().

vio_dev->unit_address only holds 32 bits of information anyways, so no
need for 64-bit field.

From Barry K. Nathan

This fixes the ulblp transfer length code, which would otherwise skip
the final iteration and lead to incomplete printer output (and us
waiting forever for the transfer to complete)

into kernel.bkbits.net:/home/davem/sparc-2.6

into nuts.davemloft.net:/disk1/BK/sparc-2.6

into debian.org:/usr/src/kernel/sparc-2.6

Here's a much better version.  The real bug was of course that the
get_user/put_user for data_direction were transposed.

This patch also fixes the translation of "quiet" and "timeout" to be
safer and clearer (instead of memcpying them as two ints and knowing
that "quiet" is first and assuming that "compat_int_t" is always the
same as the native "int", the code now handles them explicitly).

We should use "access_ok()" and __get_user/__put_user to generate better
code for this, but it's not performance-critical, so we don't care.
Some other day, perhaps.

Previous Kconfig change was incorrect, this is a better one.

On Fri, 27 Feb 2004, Linus Torvalds wrote:
> Ok, as usual, there was a lot of stuff for the -rc1, but as seems to be
> more and more true it is mainly in the "periphery".
>
> Andrew Morton:
>   o m68k: Amiga Hydra Ethernet new driver model

This part of the patch seems to have been lost (root_hydra_dev is no more):

The SA1111 appears to re-trigger an interrupt immediately following
a successful interrupt, despite no interrupt actually pending.

in our error handling thread callback.

This also exports scsi_finish_command in the SCSI layer.

Thanks much to James Bottomley and his patience, as this solution
was figured out.

Update ICS IDE driver with ide-dma and bio changes.

into kernel.bkbits.net:/home/davem/net-2.6

Maurice van der Stee noted that he got an oops on a HPFS filesystem when
saving an edited file..

  <stares at the code>
  <blinks>
  <wonders whereTF do we assign hpfs1_i and hpfs2_i if both inodes are non-NULL>
  <finds the patch in question>
  <stares at jgarzik>

This fixes it.  That, BTW, means that *nobody* had ever tried to use
hpfs r/w since 2.5.3-pre3.

From Hans Ulrich Niedermann

All sound documentation with the exception of the OSS rme96xx
documentation is under Documentation/sound/{alsa,oss}.

Move the rme966xx docs, and fix the Kconfig comments

P4 oprofile needs cpu_sibling_map and smp_num_siblings, the later
was not exported

into ppc970.osdl.org:/home/torvalds/v2.5/linux

The pdflush kernel threads can lead to an unbounded chain of thread-creations
which can overflow the kernel stacks because we didn't uses to reset
the stack on kernel thread-creation.  We do now.

Reported by Andreas Schwab, tracked down with help from Keith Owens.

This patch changes the I/O SAPIC code to cache the low 32 bits of the mask
word in kernel memory.  This greatly speeds up mask_irq() and unmask_irq().

Normally, these operations are not on the speed-critical path of the
kernel but with certain devices drivers (including users-level device-
drivers) they can become performance-critical.

        - fix bug in pfm_unload(), not allowed when not on correct CPU for
          system-wide
        - some perf/cleanup in overflow handler
        - fix reset_pmds to be on a per PMD basis on counter overflow rather
          than global
        - remove timing debug code on messages
        - no kernel info leak on PFM_END_MSG
        - remove double-store on reg_flags for pfm_write_pmcs, pfm_write_pmds
        - on restart reset_pmds is 0 by default
        - cleanup useless macros
        - cleanup some debug prints
        - added ability to remove debug code
        - streamlined sys_perfmonctl(), pfm_read_pmds(), pfm_write_*()
        - added current->tgid to default format sample header by using one
          reserved field

The problem was that IDE-disks on machines with IDE harddisks,
memory above 4GB and no hardware I/O TLB would go BUG_ON() in
blk_queue_bounce_limit() because the IDE-controller could only
address 4GB of memory and that was much less than BLK_BOUNCE_ISA
(which is equal to ISA_DMA_THRESHOLD).  Note that the problem does
NOT trigger with CD-ROMs, which always uses the software I/O TLB
(and hence bounce-buffers) on such machines.

The best fix seems to be to simply lower ISA_DMA_THRESHOLD to 4GB-1 since
that will ensure that the IDE block layer allocates memory with GFP_DMA,
which will minimize bounce buffering.

Jeff, here is a cleaned up version of the 2.6.3 hp100 isa (non-modular) probing.
It needs to check for the board being present before reading the signature.

into ppc970.osdl.org:/home/torvalds/v2.5/linux

Allow command line partitioning for use with Integrator flash.

Tighten AFS partition checks:
- check footer checksum.
- check that the image name is NUL terminated.
- return error from mtd->read intact.
This prevents us finding bogus AFS partitions.