into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-kobject

This a multi-pronged attack aimed at exploiting the kobject infrastructure mor. 

- Remove bus_driver_list, in favor of list in bus_subys.

- Remove bus_for_each_* and driver_for_each_dev(). They're not being used by
anyone, have questionable locking semantics, and really don't provide that
much use, as the function returns once the callback fails, with no indication
of where it failed. Forget them, at least for now. 

- Make sure that we return success from bus_match() if device matches, but
doesn't have a probe method.

- Remove extraneous get_{device,driver}s from bus routines that are serialized
by the bus's rwsem. bus_{add,remove}_{device,driver} all take the rwsem, so there 
is no way we can get a non-existant object when in those functions.

- Use the rwsem in the struct subsystem the bus has embedded in it, and kill the
separate one in struct bus_type.

- Move bulk of driver_register() into bus_add_driver(), which holds the bus's
rwsem during the entirety. this will prevent the driver from being unloaded while
it's being registered, and two drivers with the same name getting registered
at the same time. 

- Ditto for driver_unregister() and bus_remove_driver().

- Add driver_release() method for the driver bus driver subsystems. (Explained later)

- Use only the refcounts in the buses' kobjects, and kill the one in struct bus_type.

- Kill struct bus_type::present and struct device_driver::present. These didn't 
work out the way we intended them to. The idea was to not let a user obtain a 
rerference count to the object if it was in the process of being unregistered. 
All the code paths should be fixed now such that their registration is protected with
a semaphore, so no partially initialized objects can be removed, and enough 
infrastructure is moved to the kobject model so that once the object is publically
visible, it should be usable by other sources.

- Add a bus_sem to serialize bus registration and unregistration. 

- Add struct device_driver::unload_sem to prevent unloading of drivers 
with a positive reference count.

The driver model has always had a bug that would allow a driver with a 
positive reference count to be unloaded. It would decrement the reference 
count and return, letting the module be unloaded, without accounting for 
the other users of the object. This has been discussed many times, though 
never resolved cleanly. This should fix the problem in the simplest manner.

struct device_driver gets unload_sem, which is initialized to _locked_. When
the reference count for the driver reaches 0, the semaphore is unlocked.
driver_unregister() blocks on acquiring this lock before it exits. In the 
normal case that driver_unregister() drops the last reference to the driver,
the lock will be acquired immediately, and the module will unload. 

In the case that someone else is using the driver object, driver_unregister()
will not be able to acquire the lock, since the refcount has not reached 0,
and the lock has not been released. 

This means that rmmod(8) will block while drivers' sysfs files are open. 
There are no sysfs files for drivers yet, but note this when they do have 
some.

- Don't do extra dget() when creating symlink. 
This is a long-standing bug with a simple and obvious fix. We were doing 
an extra dget() on the dentry after d_instantiate(). It only gets decremented
once on removal, so the dentry was never really going away, and the directory
wasn't, either. 

- Use simple_unlink() instead of sysfs_unlink().

- Use simple_rmdir() instead of our own, unrolled, version. 

- Remove MODULE_LICENSE(), since it's always in the kernel. 

into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-kobject

into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-kobject

into home.transmeta.com:/home/torvalds/v2.5/linux

into mulgrave.(none):/home/jejb/BK/scsi-for-linus-2.5

On Sun, Nov 17, 2002 at 11:54:49PM +0100, Christoph Hellwig wrote:

> --- 1.46/drivers/scsi/scsi_lib.c	Thu Nov 14 18:09:17 2002
> +++ edited/drivers/scsi/scsi_lib.c	Sun Nov 17 21:37:05 2002
> @@ -7,50 +7,18 @@
>   *                        of people at Linux Expo.
>   */
>
> -/*
> - * The fundamental purpose of this file is to contain a library of utility
> - * routines that can be used by low-level drivers.   Ultimately the idea
> - * is that there should be a sufficiently rich number of functions that it
> - * would be possible for a driver author to fashion a queueing function for
> - * a low-level driver if they wished.   Note however that this file also
> - * contains the "default" versions of these functions, as we don't want to
> - * go through and retrofit queueing functions into all 30 some-odd drivers.
> - */
> -
> -#include <linux/module.h>
> -
> -#include <linux/sched.h>
> -#include <linux/timer.h>
>  #include <linux/string.h>
>  #include <linux/slab.h>
>  #include <linux/bio.h>
> -#include <linux/ioport.h>
>  #include <linux/kernel.h>
> -#include <linux/stat.h>
>  #include <linux/blk.h>
> -#include <linux/interrupt.h>
> -#include <linux/delay.h>
> -#include <linux/smp_lock.h>
>  #include <linux/completion.h>
>
> -
> -#define __KERNEL_SYSCALLS__
> -
> -#include <linux/unistd.h>

I had to add back the smp_lock.h include to compile with CONFIG_PREEMPT,
as kernel_locked was not defined and is used by in_atomic().

Patch against the latest scsi-misc-2.5:

Fix is to only plug on prep deferral if the device queue is empty
(otherwise we can rely on returning I/O to restart the queue)

* Fix compile warning in matroxfb when only 8bpp support is enabled.
* Set memory type correctly on Matrox G400.

* Executable files on ncpfs are marked by combination of SHARED and
  SYSTEM attribute, not by SYSTEM attribute alone.

  After this change gcc output is really marked executable on ncpfs.

lcall7 and lcall27 code paths are almost identical, except one
constant. This code merges these two paths together, by moving
constant to the beginning of function.

It is possible to eliminate even more of lcall7 and lcall27
code paths, but at cost of splitting SAVE_ALL into two halves,
and I do not want to do that. But if you think that it is worth
of effort, I can save 16 more bytes, but at cost of speed.

Side effects of merge is that now stack is addressed relative
to %ebx instead of relative to %esp, so generated code is
shorter and faster.

- Use PCI_BUS_NUM_RESOURCES instead of hardcoded `4' in
  pci_find_parent_resource;
- clean up pci_claim_resource() and make it a bit more informative
  on errors;
- pdev_sort_resources() must be __devinit, as it's called from
  pbus_assign_resources_sorted(), which is __devinit now;
- fix one remaining dev->name in debugging printk.

The detection of subtractive decoding bridges is broken: `class' variable
doesn't contain ProgIf byte at this point, I should check `dev->class'
instead.

This fixes resource allocation problems on certain docking stations.

The dev_t argument of sys_mknod is passed to vfs_mknod, and is then cast
to int when foo_mknod is called, and is subsequently very often cast
back to dev_t.  (For example, minix_mknod() calls minix_set_inode() that
takes a dev_t.)

This is a cleanup that avoids this back-and-forth casting by giving
foo_mknod a prototype with dev_t.  In most cases now the dev_t is
transmitted untouched until init_special_inode.

It also makes the two routines hugetlbfs_get_inode() and
shmem_get_inode() static.

Fixes miscalculation of required module size due to alignment issues of
first section after common, and also doesn't think that no init section
is an allocation failure.

Patch from Adam Richter.  I have a nicer solution based on aliases, but
it requires coordination with USB, PCI and PCMCIA maintainers, which is
taking time.

This restores the old code in the meantime: one week without this is too
long for people who need it.

into home.transmeta.com:/home/torvalds/v2.5/linux

This removes Rules.make inclusions, makes make clean work properly,
removes EXTRA_TARGETS where not needed, and fixes a couple of compile
warnings in the boot wrappers where <linux/string.h> wasn't included.

This solves some mutual inclusion problems.

Now atomic.h defines the smp_mb__* macros completely itself without
needing the smp_mb definition from <asm/system.h>

into home.transmeta.com:/home/torvalds/v2.5/linux

Also use strsep in ibmmca.c, as strtok is gone from the kernel.

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

parse_interface allocates the incorrect storage size for additional
altsettings (new buffer) leading to a BUG being triggered in
mm/slab.c:1453 when we do the memcpy from the old buffer to the new
buffer (writing beyond new buffer).
Patch appended, tested with an OV511 on an Intel PIIX4

Al pointed out that the current name of get_lease is extremely confusing
and I agree.

This (a) renames it to break_lease and (b) fixes a bug noticed by Dave
Hansen which could cause a NULL pointer dereference under high load.

into kroah.com:/home/linux/linux/BK/gregkh-2.5

This makes the driver model core (for devices) exploit the kobject
infrastructure more and make the resulting code quite a bit simpler. 

For one, device_register() mimmicks kobject_register() in that it now only
calls device_initialize() and device_add() back to back. Similarly, 
device_unregister() calls device_del() and put_device() consecutively. 

device_del() no longer removes and frees the device, it only removes them.
It also removes the devices from the global and sibling lists. This was 
previously done by device_put(), but moved here to be symmetrical with 
device_add().

The device's parent is now only incremented in device_add() and decremented
in device_del(), fixing a bug in which the parent's refcount was incremented
twice. 

Because of these simplifications, the core can easily be converted to use 
the kobject reference counting infrastructure. get_device() now simply 
forwards the call to kobject_get() and ditto for put_device(). 
device_release() is implemented to handle the freeing of devices once their
reference count reaches 0.

Since we're using the kobject refcounting model, we no longer need the 
checking or setting of the device state field, so it has been removed. 
The only users of it were the power routines. In those, it is implicit that
we have a valid device, since we've already taken device_sem, and we're 
walking the list (all modifications are protected by device_sem).

struct device::lock, and the helpers to lock/unlock have been removed. No
one has ever used them, and no one is likely to use them.

 - removed a few #ifdefs in the main code
 - cleaned up the failure logic in initialization.

On Tue, Nov 19, 2002 at 10:45:25AM +1100, Douglas Gilbert wrote:
> That directory (and all who sail in her, e.g. /proc/scsi/scsi)
> seems to have disappeared. When the scsi_debug module is
> loaded a /proc/scsi_debug/0 entry appears (that used to be
> /proc/scsi/scsi_debug/0).
>
> Doug Gilbert

It looks like the merge of Doug and Christoph's code dropped two calls
(unless the exit devfs_unregister was supposed to be removed).

Here's a patch for the addition of scsi_init_procfs, devfs_mk_dir and
bus_unregister calls, and a small reordering so calls in exit_scsi
match the reverse of those in init_scsi.

The scsi_debug version in lk 2.5.48 is the second last one
I sent to this list. So this patch includes the changes from
the last one I sent:
   - fix "in use" counting [hch]
   - clean up bios_param() code
It also merges a sysfs re-organisation from Mike Anderson.

On returning I/O, need to unplug the queue before we call the queue_fn. 
This fixes a problem in 2.5.48 where the aic7xxx driver hangs under e2fsck.

Some of that stuff might have been right for 2.4, but..

(and btw, scsi_lib is pretty misleading, what about reusing scsi_queue.c?)

Currently allocation and freeing of struct scsi_device is a mess.
We have two nice functions in scsi_scan.c (scsi_allocate_sdev/
scsi_free_sdev) that are the right interfaces to deal with it, so I moved
them to scsi and made them non-static.  I've changed all functions
allocation freeing them to use it.

*grr* - silly typedefs..