kernel weekly news – 29.10.2011

Posted: October 29, 2011 in kernel

Howdy, everyone! Here’s the news!

-Arnaldo Carvalho de Melo has again some perf/core improvements:

Arnaldo Carvalho de Melo (2):
perf ui browser: Use libslang to read keys
perf hists browser: Refuse ‘a’ hotkey on non symbolic views

Jiri Olsa (1):
perf tools: Fix tracing info recording

Ricardo Ribalda Delgado (1):
perf symbols: Increase symbol KSYM_NAME_LEN size

-Konrad Rzeszutek Wilk has various fixes/cleanups for Xen, and…
undoubtedly the news of the week is the release of Linux 3.1, as
announced by Linus Torvalds:

As promised, the kernel summit has started, and Linux-3.1 is out. The
(small) shortlog of changes since -rc10 are appended, we have mostly
some sparc and networking changes, along with some radeon and intel
iommu fixes (mostly for largepages and integrated graphics issues).
Most people probably will not notice the changes.

One big change from -rc10 is that there are tar-balls and patches, so
if you aren’t a git user (why?) you can download it now in a
traditional format. On of the things to note is that the files are now
signed by my gpg key, and it’s the *uncompressed* version that the
signature is for.

And of course, this means that the merge window for 3.2 is open. I’ll
do some merging during the KS, but probably most when I get back home
– but you can still send me the pull request, even if I may not
necessarily pull it for a few days.

NOTE! Because the -rc series was longer than usual, and as a result
linux-next is bigger than usual, I’m going to be much more of a
stickler for “has your patch series been in linux-next” than usual. If
I get a big pull request for things that I can’t find in my linux-next
branch, I will simply not pull it – we have enough code that has gone
through the proper channels as it is, and we don’t need anything

Another thing worth mentioning is that I really want the pull request
to be validated some way. With the small changes late in the -rc
series, I could afford to spend the time to look at commits and try to
verify them, but with the merge window (and the 11k commits or so that
I saw pending in the last linux-next tree), that just isn’t

So use or some other host that I can trust is really you.

-Geert Uytterhoeven has m68k updates for 3.2, while Greg Ungerer has
m68knommu fixes, and Greg Kroah Hartman announces 3.0.8. Also
Steven Whitehouse has GFS2 changes (not yet @, and
Nicholas A. Bellinger has target updates for 3.2-rc1:

The following is the first round of target updates for v3.2-rc1. It’s
available directly from here:

git:// for-next

What can I say…? Christoph really outdid himself this round with a
long list of cleanups culminating with the conversion to push the core
completion path out of transport_processing_thread() and into a separate
target workqueue. There are still a number of performance improvements
in the works, most of which at this point will end up being v3.3 items.
Also included are a handful of bugfixes reported by various folks that
are marked for, and other various improvements +
removal of legacy code.

The bulk of this series has been posted for review in the last weeks

[PATCH 00/37] target: Updates for v3.2-rc1 (round one)

Unfortunately this code has only gotten minimal linux-next testing as
the target-pending.git tree did not return to until last
week. I’ve been able to verify it does not conflict with linux-next
from 20111014, and has been getting some testing in lio-core.git and I’m
content enough for an merge into v3.2-rc1.

For the second round of v3.2-rc1 changes, a new ib_srpt module will be
merged once Roland is happy with the recent review cleanups. There is
still some outstanding non-critical breakage with last rounds conversion
to use errno.h in certain areas that still needs to be addressed, as
well as a active I/O shutdown refactoring that I’m also targeting for

-Jiri Kosina has HID and APM fixes, Greg KH has driver-core and USB
updates for 3.2, Mark Brown has regmap updates for 3.2, as

Please pull these updates for the regmap API for 3.2. In terms of the
core API the major features for the API itself are:

– Porting over of the ASoC register cache infrastructure so it can be
shared with more subsystems.
– Support for a wider range of SPI register formats.
– Debugfs and tracepoint code to support driver development.

plus a bunch of driver updates converting drivers over to the API which
depend on the core changes – the diffstat for the wm831x driver isn’t
great because it adds descriptions of the registers to make the cache
and debug infrastructure much more useful, the amount of code has shrunk
but we’ve now got more data.

There’s one patch converting the wm831x driver to use the register cache
code which hasn’t hit -next due to Stephen Rothwell’s holiday – it a one
line patch which has been percolating in my tree for a while and all the
users of this driver in tree are boards I maintain.

As the pull request is against 3.1-rc1 there’s a couple of small bugfix
patches in here which you’ve already pulled into 3.1, I’m not sure if
there’s a sensible way to deal with show those in the pull request other
than doing a merge up of 3.1?

The following changes since commit 322a8b034003c0d46d39af85bf24fee27b902f48:

Linux 3.1-rc1 (2011-08-07 18:23:30 -0700)

are available in the git repository at:
git:// for-linus

and David Miller has networking updates:

The most visible thing in here is the reorganization of the drivers
under drivers/net.

Mostly it’s moving the various ethernet drivers to paths of the form

There is almost nothing left directly under drivers/net/

Thanks to Jeff Kirsher for doing this work.

I have to say I was really impressed with GIT’s rename detection. For
example, when I’d pull bug fixes from the ‘net’ tree into this
‘net-next’ tree, GIT would figure it out and cleanly merge in the
fixes even though the file names were completely different.

I’ll also note that I had to bump GIT’s diff.renamelimit to about 1000
to generate this pull request.

Other highlights:

1) Ian Campbell is going through the networking abstracting away
accesses to SKB page frags so we can cheaply do interesting things
with page referencing in the future.

2) Eric Dumazet is increasing the accuracy of skb->truesize’s value
on incoming SKBs from devices so that we can avoid unexpectedly
running out of memory before we hit the individual socket and the
global protocol socket memory limits.

I think that, without the driver renames, this is actually a smaller
networking pull request than usual for the merge window.

Please pull, thanks a lot.

The following changes since commit 1bc67188c3843b8e16caaa8624beeb0e2823c1f8:

Merge branch ‘for-linus’ of (2011-10-25 10:18:27 +0200)

are available in the git repository at:

git:// master

-Trond Myklebust has lots of updates to the NFS (client) tree,
Rafael J. Wysocki updates the power management tree

* Runtime PM tracepoints (replacing not very useful debug printks) and fixes
from Ming Lei.

* Introduction of devfreq by MyungJoo Ham (special thanks to Mike Turquette
for reviewing all of the versions of this patchset).

* Introduction of device PM QoS from Jean Pihet.

* Generic PM domains core code refinements and extensions.

* SH7372 power management changes from Magnus Damm.

* Threaded hibernate image saving from Bojan Smojver.

* New entries for the ACPI suspend blacklist.

* Assorted fixes.

Boaz Harrosh has exofs updates, as seen below,

The biggest highlight of this Merge is support for RAID5 over objects
with full support for the pnfs-objects-ietf-standard layout. (This is
a special RAID5 over objects which is equivalent to RAID5 over files
much more then RAID5 over block devices).

Part 2: Of this merge is dependent on this inclusion and the bits
in Trond’s NFS tree. For the fs/nfs/objlayout and will be pushed
through Trond, once both hit mainline.

and Guenter Roeck has hwmon updates, for 3.2 of course.

-Neil Brown has md updates for 3.2:

Hi Linus,
please pull following md updates for 3.2
Most of them are just code cleanups and minor bug fixes.
There is one fix for a RAID5 bug which could lead to data corruption
due to reading old data from a device that is marked as faulty. It is a
small race as the drive is normally removed from consideration very soon
after being marked faulty. This patch is flagged for inclusion in -stable.
It is commit 355840e7a7e56bb.

This series conflicts with the patch
block: remove support for bio remapping from ->make_request
from Christoph Hellwig in the block tree.
That patch changes ‘make_request’ functions to return void. I have patches
which change the types of args to some of those functions from a typedef to
a ‘struct XXX’.
The conflict is fairly easy to resolve (“wiggle” does it perfectly), or you
could pull my “for-next” branch (85d7a4e56fbe09c1) instead which includes
block-3.1 and resolves the conflict.

-Ingo Molnar has RCU, core, IRQ, perf and scheduler updates for 3.2, he also
worked on some timers-core, apic, cleanups and asm on x86, Pekka Enberg
has slab updates, also for 3.2, and the same Ingo has x86/{cpu,cpufeature,
goede,hyperv,microcode,mm,rdrand,spinlock,uv,vdso,} updates for 3.2 .

-Arnaldo Carvalho de Melo has perf/core fixes:

Please consider pulling from:

git:// perf/core

It _almost_ is libnewt free, got preempted now, will continue later.

Using libnewt ended up being inflexible and since I had to implement
the ui_browser to work directly with the existing perf hist_entry abstractions,
namely callchains, I ended up using it for menus and for the annotate browser.

It got less and less used and finally handling SIGWINCH made me decide
to work towards using just libslang.

The abstractions that are being put in place should make it not that
difficult to have a GUI at some point.

-Jeff Garzik has updates for libata :

1) many minor PATA fixes and cleanups from Bart Z

2) pata_legacy has supported QDI PATA for a while, so we may
eliminate pata_qdi driver as duplicate. (Bart Z, acked by Alan)

3) add power management support to several PATA drivers (Bart Z)

4) other minor PATA, SATA driver fixes and cleanups (various)

5) fix ATAPI device + Sandybridge issue (Intel, Canonical)

6) make pata_of_platform work again on !PPC — though note we hope to replace
this with pata_platform + DT (patch by Grant L exists)

7) core bug fix from Tejun: garbage data could be transferred to an
ATAPI device as padding, Presumably, as this is padding, the bytes are
irrelevant — but at least one nutty device wants padding == zeroes.

administrivia notes:

1) This entire pull request is GPG-signed by a new GPG key 01782B2F
(, AF8B E07C 7049 F3A2 6B23 9D53 25B3 0832 0178 2B2F),
created for new interactions. However, not being at KS, I’ve
only gotten Andre Hedrick to sign this new key so far. In addition to keyserver, my pubkey is published on my personal website at

2) The GPG wrapping causes all ‘^-‘ lines to become ‘^- -‘. This does NOT
reflect the true ‘git diff’ output, of course.

For this reason, I am tempted to modify my new, GPG-signed pull request
template as follows, for subsequent pull requests:


pull req description

“Please pull from $branch\n$url $branch”

“Top of tree is $sha1_commit”



That ensures that the critical part — sha1 commit for top of tree —
is GPG signed, while the diff will be outside the signed area and therefore
not mangled by wrapping.

-Dmitry Torokhov issues a big update set for the input tree (aimed at
3.2-rc0), Dan Magenheimer has mm updates/fixes and Dave Airlie updates
the drm tree.

-That’s it for me, see you next week!


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s