-We start this week’s news with Tony Lindgren’s OMAP fixes aimed
at -rc5; Ingo Molnar has the usual updates : x86, timer and perf
fixes, James Bottomley announces a few SCSI fixes for -rc5 and
Tejun Heo has wq fixes:
Please pull from the following branch to receive fix for a deadlock which can happen between rescuers and stop_machine() while bringing down CPUs on non-preemptive kernels. The problem could be half-reliably reproduced by Thilo-Alexander while the system is entering suspend. The branch also contains another commit which documents workqueue debugging tricks. The pull will cause conflict against 25985edced (Fix common misspellings). The fixed misspelling unfortunately was in the comment which the fix replaces. The conflict can be resolved by simply taking the version from the fix commit. git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git fixes-2.6.39
-Willy Tarreau announces 22.214.171.124 – ftp://ftp.all.kernel.org/pub/linux/kernel/v2.6/longterm/v2.6.27/ -,
Tejun Heo has a pull request regarding x86/mm, namely unify 32/64-bit NUMA init
paths , Greg Kroah Hartman announces the release of 126.96.36.199 and David Miller
announces changes to the networking tree :
The commit count here is not as low as I would have liked, but it is the usual smattering of one-liners fixing problems people are actually running into. Considering the fact that the last networking pull request was nearly 2 weeks ago, the hit count isn't that bad. Along with the straight-out fixes, there are a few reverts for changes that caused clear regressions. Anyways: 1) Revert veth change, breaks configuration of veth devices, from Jiri Pirko. 2) WoL by default isn't done properly in tg3 driver, from Rafael J. Wysocki. 3) fs_enet driver doesn't configure multicast into the chip properly, from Andrea Galbusera. 4) Fix QoS config send halting TX in iwlwifi, from Daniel Halperin. 5) atl1c manages async work events using non-atomic bit field, use atomic bitops to fix these races, from Tim Gardner. 6) Revert bridging change which broke 802.3ad bonding inside of bridge. 7) Removing a net driver can wedge in netconsole due to recursive locking, fix from Neil Horman. 8) bnx2x blindly dereferences packets as if they were all TCP, fix to check protocol type properly and handle UDP correctly as well. From Eric Dumazet and Vladislav Zolotarov. 9) Fix r8169 to not request firmware before userspace is even present, from Francois Romieu. 10) Blackhole routes need to provide ->cow_metrics() op otherwise we OOPS, from Held Benhard. 11) IPSEC replay window handling fixes from Steffen Klassert. 12) usbnet can wedge on resume because a pending interrupt URB gets lost, resubmit it on resume. From Paul Stewart. 13) Pending timer on remove results in OOPS in bnx2 driver, fix from Neil Horman. 14) ftmac100 does usleep_range() in non-sleepable contexts, fix to use udelay() for now. From Adam Jaremko. 15) AF_UNIX sockets use a dummy packet send in order to implement connect() and accept(). We want to completely ignore these pseudo-packets in recvmsg() otherwise we can OOPS. Fix from Eric W. Biederman. 16) Every net namespace creation emits an annoying log message from the fib_trie code, kill it. From Alexey Dobriyan. 17) Invalid ioctl commands should emit -ENOTTY not -EINVAL. From Lifeng Sun. 18) smsc95xx driver checks BMCR register bit improperly, it's actually using a register address (which is zero and thus the test never passes) :-) Sharp eyed fix from Rabin Vincent. 19) Calls to register_net_sysctl_table() should be consistently paired with unregister_net_sysctl_table(). Fix broken cases, from Lucian Adrian Grijincu. 20) RAW CAN socket release can OOPS due to missing 'sk' check, from Oliver Hartkopp. The BCM CAN code has an identical bug, also fixed here, as was originally reported by Dave Jones. These bugs were uncovered thanks to Dave's syscall stresser. 21) SKB use after free fixes in iwlwifi and iwl4965, from Stanislaw Gruszka. 22) udp6_ufo_fragment() mis-calculates SKB headroom, from Shan Wei. 23) A conversion of how the RTAX_HOPLIMIT mertric is handled in ipv6 missed one case, resulting in addrconf routes always using "255" as the hop limit. Fix from Thomas Egerer. 24) Missing counter bump in be2net driver results in temperature sense happening way too often, from Ivan Vecera. 25) agg_device_up() in the bonding code uses the wrong method to find the aggregator of the bond, it should use agg->lag_ports->slave instead of agg->slave. Fix from Jiri Bohac. 26) Bluetooth HCI connect code does one refcount grab too many, fix from Ville Tervo. Please pull, thanks a lot!
-Chris Ball announces some MMC fixes aimed at 2.6.39-rc6, Ingo Molnar
, as usual, updates x86 and irq, Dave Airlie has updates for the drm tree,
James Morris updates the security-testing tree, Frederic Weisbecker
has hw_breakpoint fixes and …..
-Linus Torvalds announces 2.6.39-rc6, with the following immortal words:
Most of the stuff here no normal user will ever really notice (like the ARM kprobes series), and most of the rest is the usual one-liner random driver stuff. But we had one really interesting bug that showed itself as an RCU failure on UP, which turned out to be due to a scheduler problem, which in turn turned out to be a timer initialization issue. That was interesting and rather obscure. But most of it is just pretty random. Some media updates, some NFS work, some drm fixes, documentation, etc. Pretty small, and the bugs have been pretty obscure on the whole. We're still chasing some stuff down, but I think we're ok for -rc6. This isn't going to be the final -rc, but it doesn't seem to be in bad shape, and people who have posted regression reports please check them again, and people who haven't, please do give it a test.
-Mauro Carvalho Chehab has updates for v4l/dvb (-rc7), Sage Weil
has ceph fixes , also for -rc7, usb fixes are issued by Greg Kroah
Hartman, John Stultz announces rtc fixes for 2.6.39 and
…this week is over, so I wish you a fun weekend !