into delerium.codemonkey.org.uk:/mnt/nfs/neologic/bar/src/kernel/2.6/trees/agpgart

From: Matt Domsch.
The E7205 doesn't have an AGP header, so printing this message is
pretty much useless.  Also make it KERN_WARNING as well, as it's
not really worthy of a KERN_ERR
 

(the bios might not have done that).

From: Arjan van de Ven

This option only worked for the amd64 driver.
On every other driver, the only thing it did was
make it not printk the banner on startup.

This is horribly broken due to a jiffy wrap bug, we never get out of
the while loop, preventing booting on a kernel with this driver
compiled in. (See https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=124495)

The warning message there has never been reported in any bug report
that google can find, so I'm of the opinion that this driver is trying
to be too clever for its own good.  Rip out the jiffies logic completely,
it should be totally unnecessary.

Assigning '*wdata' with a dynamic initializer creates a
temporary structure copy on the stack, and then the final
data is initialized with a "memcpy()".

As a result, these NFS functions use more than 800 bytes
of stack-space.

Changing the code to just do a memset followed by explicit
initialization of the non-zero member variables takes the
stack usage down to 36 bytes. 

Here is a patch that does exactly that.

Removed one (harmless) direct user pointer access.

Also use "__force" when forcing the user address to "unsigned long"
for the address limit check.

Direct sscanf() on userland buffer in procfs ->write_proc().  Fixed.

Everything straigthforward and takes care of almost all sparse warnings
in fs/reiserfs/*.

Several places in arch/i386/kernel were still un-annotated - they didn't
trip copy_.._user(), so they stayed alive.  Caught and killed.

That's not all - there's a big cluster of them in vm86.c, but I haven't
looked into that code yet.

A bunch of "return in void function" dealt with, removed bogus extern
from definition of __load_new_mm_context(), sanitized fscking ugly
CROSS_64K() helper (even if sparse doesn't segfault on that anymore,
it's _still_ fscking ugly).

	... and that's where I'd found do_utimes() crap before I went
grepping ;-)

Trivial stuff in arch/alpha annotated; in two cases of
osf_proplist_syscall() dereferencing userland pointers replaced with
get_user

#if where #ifdef should've been

Long constants marked as such

Copying primitives annotated.

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

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

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

Fixed the type of SIOCGIFCONF callback; inet instance was already
correctly annotated, decnet one was _not_.  Moreover, decnet callback
needed fixing - dereferencing userland address.

Made ->write_proc in tty_driver suitable for procfs write callback

n_tty annotated, ldisc_struct ->read() and ->write() got __user
on their buf argument.

vt, vt_ioctl, consolemap and selection annotated, struct unimapdesc
and struct consolefontdesc got __user on their pointer members.

tty_ioctl.c annotation

tty_io.c annotated

That "u64" in there is actually struct iocb __user *.  Changed
prototype, killed casts in body and caller...

return void in floppy.h and "conditionally long" constants in page.h.

Would be useful to teach as(1) about C notation for long...

Long constants marked as such.

#if where #ifdef should've been.  Fixed.