On Thu, Nov 04, 2004 at 12:52:38PM -0800, Greg KH wrote:
> Hi,
>
> I get the following BUG in the sysfs code when I do:
> 	- plug in a usb-serial device.
> 	- open the port with 'cat /dev/ttyUSB0'
> 	- unplug the device.
> 	- stop the 'cat' process with control-C
>
> This used to work just fine before your big sysfs changes.

There is a similar problem reported by s390 people where we see parent
kobject (directory) going away before child kobject (sub-directory). It
seems kobject code is able to handle this, but not the sysfs. What could
be happening that in sysfs_remove_dir() of parent directory, we try to
remove its contents. It works well with the regular files as it is the
final removal for sysfs_dirent corresponding to the files. But in case
of sub-directory we are doing an extra sysfs_put().  Once while removing
parent and the other one being the one from when sysfs_remove_dir() is
called for the child.

The following patch worked for the s390 people, I hope same will work in
this case also.


o Do not remove sysfs_dirents corresponding to the sub-directory in
  sysfs_remove_dir(). They will be removed in the sysfs_remove_dir() call
  for the specific sub-directory.
Signed-off-by: Maneesh Soni <maneesh@in.ibm.com>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

into kroah.com:/home/greg/linux/BK/driver-2.6

kfifo_alloc tries to round up the buffer size to the next power of two.

But it accidently uses the original size when calling kfifo_init,
which will BUG.
Acked-by: Stelian Pop <stelian@popies.net>
Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Attached patch adds a MAINTAINER entry for the kernel event layer.
Signed-Off-By: Robert Love <rml@novell.com>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

Looks like kobject_uevent_init is executed before netlink_proto_init and
consequently always fails.  Not cool.

Attached patch switches the initialization over from core_initcall (init
level 1) to postcore_initcall (init level 2).  Netlink's initialization
is done in core_initcall, so this should fix the problem.  We should be
fine waiting until postcore_initcall.

Also a couple white space changes mixed in, because I am anal.
Signed-Off-By: Robert Love <rml@novell.com>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

Thanks to Serge Hallyn <serue@us.ibm.com> for pointing this out.
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

This clarifies more of the x86 caller/callee stack ownership
issues by making the exception and interrupt handler assembler
interfaces use register calling conventions.

System calls still use the stack.

Tested with "crashme" on UP/SMP.

 df_05_device_add_ref_fix.patch

 In device_add(), @dev wan't put'd properly when it has zero length
bus_id (error path).  Fixed.
Signed-off-by: Tejun Heo <tj@home-tj.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

 df_04_kobject_add_ref_fix.patch

In kobject_add(), @kobj wasn't put'd properly on error path.  This
patch fixes it.
Signed-off-by: Tejun Heo <tj@home-tj.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

 df_03_sysfs_release_fix.patch

Some attributes are allocated dynamically (e.g. module and device
parameters) and are usually deallocated when the assoicated kobject is
released.  So, it's not safe to access attr after putting the kobject.
Signed-off-by: Tejun Heo <tj@home-tj.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

 df_02_bus_rescan_devcies_fix.patch

 bus_rescan_devices() eventually calls device_attach() and thus
requires write locking the corresponding bus.  The original code just
called bus_for_each_dev() which only read locks the bus.  This patch
separates __bus_for_each_dev() and __bus_for_each_drv(), which don't
do locking themselves, out from the original functions and call them
with read lock in the original functions and with write lock in
bus_rescan_devices().
Signed-off-by: Tejun Heo <tj@home-tj.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

 df_01_driver_attach_comment_fix.patch

bus_match() was renamed to driver_probe_device() but the comment for
device_attach() wasn't updated.  This patch updates it.
Signed-off-by: Tejun Heo <tj@home-tj.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

Add the sysfs path of the physical device to the hotplug event of class
and block devices. This should solve the userspace issue not to know if
the device is a virtual one and the "device" symlink will never be created,
but we sit there and wait for it to show up not knowing when we should
give up.

Also the bus name is added to the hotplug event, so we don't need to
reverse lookup in the /sys/bus/* directory which bus our physical
device belongs to. This is e.g. the value matched against the BUS= key,
that may be used in an udev rule.

This is a PCI network card:
  ACTION=add
  SUBSYSTEM=net
  DEVPATH=/class/net/eth0
  PHYSDEVPATH=/devices/pci0000:00/0000:00:1e.0/0000:02:01.0
  PHYSDEVBUS=pci
  INTERFACE=eth0
  SEQNUM=827
  PATH=/sbin:/bin:/usr/sbin:/usr/bin
  HOME=/

This is a IDE CDROM:
  ACTION=add
  SUBSYSTEM=block
  DEVPATH=/block/hdc
  PHYSDEVPATH=/devices/pci0000:00/0000:00:1f.1/ide1/1.0
  PHYSDEVBUS=ide
  SEQNUM=1017
  PATH=/sbin:/bin:/usr/sbin:/usr/bin
  HOME=/

This is an USB-stick partition:
  ACTION=add
  SUBSYSTEM=block
  DEVPATH=/block/sda/sda1
  PHYSDEVPATH=/devices/pci0000:00/0000:00:1d.1/usb3/3-1/3-1:1.0/host1/target1:0:0/1:0:0:0
  PHYSDEVBUS=scsi
  SEQNUM=1032
  PATH=/sbin:/bin:/usr/sbin:/usr/bin
  HOME=/
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

We need to open the gap between congestion on/off a little bit, or we
risk burning many cycles continually putting processes on a wait queue
only to wake them up again immediately. This was observed with CFQ at
least, which showed way excessive sys time.

Patch is from Arjan.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Patch is from Daniel Drake <dsd@gentoo.org>
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Daniel Drake <dsd@gentoo.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

into ppc970.osdl.org:/home/torvalds/v2.6/linux

Signed-off-by: David S. Miller <davem@davemloft.net>

Using idr_get_new_above in init_new_context lets us get rid of an
awkward init function which wasn't running early enough in boot
anyway.
Signed-off-by: Nathan Lynch <nathanl@austin.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

into ppc970.osdl.org:/home/torvalds/v2.6/linux

into ppc970.osdl.org:/home/torvalds/v2.6/linux

The routine ext2_readdir() will, when reading a directory page
returns an error, try the next page, without reporting the
error to user space. That is bad, and the patch below changes that.

In my case the filesystem was damaged, and ext2_readdir wanted
to read 60000+ pages and wrote as many error messages to syslog
("attempt to access beyond end"), not what one wants.

[no doubt a similar patch is appropriate for ext3]

This preserves 2.4.x behavior.
Signed-off-by: David S. Miller <davem@davemloft.net>

It is updated outside of locks, so...
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Collaborative work between David S. Miller and
Herbert Xu.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

bits_wanted is expanded to bytes using the wrong shift value
(when iovp_shift != PAGE_SHIFT), resulting in an explosion of
used iommu resources.

This potentially results in mistakenly running out of DMA mapping
resources when the system is under *heavy* i/o load.
Signed-off-by: Nigel Croxon <nigel.croxon@hp.com>
Signed-off by: Alex Williamson <alex.williamson@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: David S. Miller <davem@davemloft.net>

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

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Transforms old policer to use generic statistic via TCA_STATS2
and maintain backward compatibility via TCA_STATS. Adds a new
API for classifiers to invoke the dumping process.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Dumps the statistic common to all action modules via the newly
introduced TLV type TCA_ACT_STATS in tcf_action_copy_stats but
allows the corresponding module to dump its own statistic by
implementing the get_stats callback.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

The patch below does the following cleanups for the sysfs code:
- remove the unused global function sysfs_mknod
- make some structs and functions static

Please check whether this patch is correct, or whether some of the
things I made static should be used globally in the forseeable future.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

o sysfs_new_dirent to retrun 0 if kmalloc fails. Thanks to Milton Miller
  for spotting this.
Signed-off-by: Maneesh Soni <maneesh@in.ibm.com>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>

ia64_sn_plat_specific_err_print() ends up calling vfree() which
requires IPI which can deadlock if interrupts are disabled.  Do not
disable interrupts in ia64_sn_plat_specific_err_print().
Signed-off-by: Keith Owens <kaos@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Patch from Matthew Wilcox:
  Since some ia64 platforms have non-identity-mapped PCI busses, we need
  to implement pcibios_resource_to_bus and pcibios_bus_to_resource.
  Also clean up some >80 column comments and delete an unnecessary parameter
  to pcibios_fixup_device_resources().
Signed-off-by: Tony Luck <tony.luck@intel.com>

Based upon a patch from Herbert Xu:

> Hi Ingo:
> 
> Your recent fix to dev_queue_xmit moved the local_bh_disable to
> include stuff like skb_linearize and skb_checksum_help.  These
> are potentially expensive operations.  Since they don't need to
> run with preempt off, we could simply move the local_bh_enable
> up instead.

No, this breaks the normal return path.

How about this instead?
Signed-off-by: Tommy S. Christensen <tommy.christensen@tpack.net>
Signed-off-by: David S. Miller <davem@davemloft.net>