Alan Stern wrote:
> I finally got tired of my computer locking up when I tried to rmmod the
> low-level host controller driver.  It turns out the problem lies in the
> root-hub status urb code in core/hcd.c -- primarily a result of
> rh_report_status() not calling hcd_giveback_urb()...

Or in short:  your patch removes some old logic for the "automagic
interrupt transfer" special casing ... which recently started to
break that rmmod path.

With automagic, the only time an interrupt urb (like the root hub
status urb) could legitimately be given back was for unlink.  But
that unlink doesn't seem to be issued in the same way lately during
the rmmod paths.  (If they're less bizarre lately, that's good!)


> If this patch seems all right, will you please let Greg know it's okay to
> apply it?

I changed a couple minor things below ... basically (a) fixing the
issue Duncan Sands pointed out (always call completions with irqs
disabled, even if hub driver currently doesn't care), (b) better
logic to avoid retriggering the timer during shutdown, (c) not
doing del_timer_sync() while holding that lock, plus (d) a minor
linewrap fix.

This patch removes a single-line inline function and replaces it with it's
expansion everywhere.   The inline function was really only there for a
transitional step.

This patch forces the generation of a bulk-transport reset after a command
abort.  Careful reading of the bulk-only specification shows that a reset
must be done after an abort to get the host and target back into phase.

I tried compiling USB gadget support on ppc64 (why not :). Looks like Im
seeing a conflict between const and __devinitdata:

drivers/usb/gadget/net2280.c:2645: pci_ids causes a section type conflict

usb-skelteon doesn't compile after recent changes.

Supports some user-mode testing "firmware".

This is what a lot of the callers really wanted.

Reiserfs emits a warning about strchr being defined but not used.  I
finally tracked down the reason for this.  gcc - when seeing strstr(x,
"%") recognized that the second parameter is a char, and therefore uses
strchr instead of strstr.  The workaround to avoid the warning is to
replace the call to strstr with strchr - which is OK.

This hides the warning, and brings us down to 6 warnings for a make
defconfig bzImage.

This always set ->sense for blk_pc_requests(), even if the actual user
doesn't care about the sense results.  This is a lot nicer than then
having to have conditional checks for it all over the place in the low-
level drivers.

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

into kernel.bkbits.net:/home/davem/net-2.5

codepath.

The bug was that the NMI stack fixup triggered even if the
debug exception had _not_ happened on the exact sysenter
entrypoint. The new version should be more robust.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

Also use kernel-doc for struct cycx_hw

. remove whitespaces
. use tabs instead of sequences of 8 spaces
. remove the wrappers for write{b,w} & friends
. align case entries with corresponding switch statement

into kernel.bkbits.net:/home/davem/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5