This patch against 2.5.75 initialises a spinlock when the structure containing
it is allocated

This patch is needed to make software suspend work with the 8139too
driver loaded.

This patch fixes another way the Rhine-I found to break down under load. It
should bring Rhine-I behavior on par with the Rhine-II.

* Add ethtool TSO, Rx/Tx csum, SG Get/Set support.

* Add TSO get/set command to ethtool interface.  Applies to both 2.4/2.5.
  Ethtool application patch sent under separate cover.

ne2k-pci leaks memory on unload. dev->priv is allocated in ethdev_init(), but
never freed. against 2.4-bk, but also applies to 2.5-bk with offset.

into redhat.com:/garz/repo/net-drivers-2.6

Previous entry said to be out of date by two years or more.

As noted by Gergely Nagy:

  "devfs_mk_cdev() first checks the mode passed to it, and if it thinks
   it is not a char device, it prints a warning and aborts.  Now, this
   printing involves the local variable `buf' (char buf[64]), which is
   not initialised at that point."

The same problem also affects devfs_mk_bdev.

Fixed thus.

davidm@hpl.hp.com writes:
  "I'm working on updating the ia64 tree with local_t etc.  One
   thing that would really help me: could you make
   asm-generic/percpu_modcopy() a macro?  The routine depends on
   cpu_possible(), but I can't including smp.h in percpu.h since
   that would lead to recusive header-file dependencies (and in my
   opinion, percpu.h should be more "primitive" than smp.h, so that
   it can be included virtually everywhere)."

A few places pre-declare "int module_init(void);" and "void
module_cleanup(void);".  Other than being obsolete, this is
unneccessary (it's in init.h anyway).

There are still about 100 places which still use the
obsolete-since-2.2 "a function named module_init() magically gets
called": this change frees us up implement that via a macro.

Author: Neil Brown

Define module_put_and_exit() and use it for nfsd/lockd

Both nfsd and lockd have threads which expect to hold a reference
to the module while the thread is running.  In order for the thread
to be able to put_module() the module before exiting, the
put_module code must be call from outside the module.

This patch provides module_put_and_exit in non-modular code which a
thread-in-a-module can call.  It also gets nfsd and lockd to use it
as appropriate.

Note that in lockd, we can __get_module in the thread itself as the
creator of the thread is waiting for the thread to startup.

In nfsd and for the 'reclaimer' threaded started by locked, we
__get_module first and put_module if the thread failed to start.

Somehow, the code which taints the kernel when rmmod -f is used got
lost.  Restore it.

These are the local_t bits for sparc64.

Other platforms such as PPC64 and Alpha (and in fact any "load locked/
store conditional" like platform other than IA64), probably want to do
something similar.

Uses local_t for module reference counts.

Introduces local_t, a type which is like atomic_t, but the
operations are only atomic from a single CPU's point of view
(ie. atomic against interrupts and softirqs). Some architectures
(eg. i386) can implement these very efficiently.

There are special operations for the case of a local operation on a
per-cpu variable (which is common), which some architectures can
implement efficiently (eg. IA64 keeps all per-cpu variables mapped
at the same address, so no address arithmetic is needed).

The generic implementation given here simply uses atomics on 32-bit
archs, three variables on others.  x86 is already implemented specially.

Richard Henderson point out a while back that linker generated symbols
should be declared as: "char _text[]" so that the compiler can't make
assumptions about them sitting in small sections, etc.

Centralize these defintions in asm/sections.h (where some already
are on x86).

The current percpu macros do not allow __get_cpu_var(foo.val1)
which makes building macros on top of them really difficult.  The
reason for this is the __per_cpu postfix appended to per-cpu
variables, designed to catch direct uses.  Prepend it instead.

tests.

From Dennis Vshivkov: 

  "whenever I was trying to create a file using only national
   characters for its name, it was always created with an
   uppercased shortname (provided that the name was not too
   long for 8.3, of course). Everything was fine with latin
   filenames, though."

The bug is that we check "buf[0]" as an unsigned character, but
"buf" is just "char *", which (depending on architecture and
compiler options) tends to be signed.

The joys of having several arches at my home lab, thanks
to parisc this time.

Adds kahlua, harmony and hal2 drivers to the build options.

Fix an assortment of unchecked copy_to_user().

Memory leak fix: hw_config is allocated before the call to sb_dsp_detect.

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