Marcel Holtmann [Wed, 14 Jan 2004 01:34:23 +0000 (02:34 +0100)]
[Bluetooth] Fix reference counting for incoming connections
The L2CAP reference counting must be increased by the RFCOMM module if an
incoming connection is accepted. If this is not done, it will be decreased
one time to often if the connection is terminated.
Krishna Kumar [Tue, 13 Jan 2004 15:26:01 +0000 (07:26 -0800)]
[XFRM]: Fix two bugs in xfrm_lookup()
- the found or allocated xfrm_states are not passed correctly to
xfrm_bundle_create (and to the subsequent frees in case of create
failing) if the first xfrm_tmpl_resolve failed and the second one
succeeded.
- error handling is wrong.
Remove duplicate/conflicting code now that netdev_register_notifier replays the
device registration events. Reorder initialization to avoid having to simplify error unwind
if /proc creation failed.
Change IPV6 to handle the new case where netdev events are replayed on registration:
* change code ordering so address configuration is ready before registration
(this was actually a bug in existing code)
* take out code that capture's existing devices, this is now done in the replay
* make notifier code local to addrconf.c so it can be have smaller scope
Marcel Holtmann [Tue, 13 Jan 2004 05:17:12 +0000 (06:17 +0100)]
[Bluetooth] Set disconnect timer for incoming ACL links
In general a Bluetooth stack that creates an ACL link should also take
care of its termination, but some Bluetooth devices think otherwise. The
problem is that the Bluetooth specification don't defines an ownership
of an ACL link and if nobody cares about it, two devices stay connected
even if there is no need. To avoid leftover ACL links the disconnect
timer must also be used for incoming connections, but it is set twice
the default disconnect timeout so the other side get enough time to
clean everything by itself.
Linus Torvalds [Thu, 8 Jan 2004 03:46:52 +0000 (19:46 -0800)]
Fix subtle fork() race that Ingo noticed.
We must not mark the process TASK_STOPPED early, because
that might allow a signal to wake it up before we actually
got to the "wake_up_forked_process()" state. Total confusion
would happen.
Make wake_up_forked_process() verify the new world order.
Dave Jones [Wed, 7 Jan 2004 14:53:58 +0000 (14:53 +0000)]
[AGPGART] Add support for Radeon IGP345M to ATI GART driver.
Also consolidate a bunch of convoluted if's into a single
"is this an r200 or r300" function.
Andrew Morton [Wed, 7 Jan 2004 13:56:59 +0000 (05:56 -0800)]
[PATCH] ppc64: Add missing section definition
From: Anton Blanchard <anton@samba.org>
For a laugh we moved the BSS and our world exploded. Turns out we zero
from __bss_start to _end. Add __bss_stop to our section definitions so
we can use it instead.
Dave Kleikamp [Wed, 7 Jan 2004 02:35:15 +0000 (18:35 -0800)]
[PATCH] don't clear i_sb
From: Christoph Hellwig <hch@lst.de>
JFS currently clears i_sb in some error pathes which can make the
core kernel OOPS because it may never be NULL. Noticed because some
IBM people try to "fix" the core kernel for it now..
Andrew Morton [Tue, 6 Jan 2004 23:52:48 +0000 (15:52 -0800)]
[PATCH] ia32 sched_clock() deadlock fix
From: Ingo Molnar <mingo@elte.hu>,
Voyager is getting odd deadlocks due to the taking of xtime_lock() in
sched_clock()->get_jiffies_64().
I had this patch queued up to fix a different deadlock, which occurs when we
relax the requirement that TSC's be synchronised across CPUs. But it will
fix James' deadlock too.
Andrew Morton [Tue, 6 Jan 2004 23:52:30 +0000 (15:52 -0800)]
[PATCH] JFS fix for NFS on little-endian systems
From: Dave Kleikamp <shaggy@austin.ibm.com>
After Jose debugged the problem down to the routine jfs_get_parent, we
were able to find the problem. I believe it only affects users of
NFS-exported JFS file systems on big-endian hardware.
The problem was a missing le32_to_cpu macro. The patch also fixes a
return code to be more consistent other implementations of get_parent.
Andrew Morton [Tue, 6 Jan 2004 23:52:21 +0000 (15:52 -0800)]
[PATCH] fix sysfs oops
From: Greg KH <greg@kroah.com>
This fixes an oops when a kobject is unregistered before it's child is.
The usb-serial devices show this bug very easily (yank out a device
while its port is opened...)
Patch was originally written by Mike Gorse <mgorse@mgorse.dhs.org>
Andrew Morton [Tue, 6 Jan 2004 23:52:12 +0000 (15:52 -0800)]
[PATCH] MSI build fixes
Add missing NR_VECTORS definition to visws and pc9800. Also, make MSI
support dependent on CONFIG_X86_IO_APIC, as it won't build without IOAPIC
support.
Adam Belay [Tue, 6 Jan 2004 12:08:21 +0000 (04:08 -0800)]
[PATCH] Fix PnP BIOS call
The recent escd fix I have made corrects a thinko in the PnPBIOS code and it
turns out that faults from calling /proc/pnp/bus/escd were probably not caused
by BIOS bugs.