We have reached the point where our mutexes are quite fine tuned
for a number of situations. This includes the use of heuristics
and optimistic spinning, based on MCS locking techniques.
Exclusive ownership of read-write semaphores are, conceptually,
just about the same as mutexes, making them close cousins. To
this end we need to make them both perform similarly, and
right now, rwsems are simply not up to it. This was discovered
by both reverting commit 4fc3f1d6 (mm/rmap, migration: Make
rmap_walk_anon() and try_to_unmap_anon() more scalable) and
similarly, converting some other mutexes (ie: i_mmap_mutex) to
rwsems. This creates a situation where users have to choose
between a rwsem and mutex taking into account this important
performance difference. Specifically, biggest difference between
both locks is when we fail to acquire a mutex in the fastpath,
optimistic spinning comes in to play and we can avoid a large
amount of unnecessary sleeping and overhead of moving tasks in
and out of wait queue. Rwsems do not have such logic.
This patch, based on the work from Tim Chen and I, adds support
for write-side optimistic spinning when the lock is contended.
It also includes support for the recently added cancelable MCS
locking for adaptive spinning. Note that is is only applicable
to the xadd method, and the spinlock rwsem variant remains intact.
Allowing optimistic spinning before putting the writer on the wait
queue reduces wait queue contention and provided greater chance
for the rwsem to get acquired. With these changes, rwsem is on par
with mutex. The performance benefits can be seen on a number of
workloads. For instance, on a 8 socket, 80 core 64bit Westmere box,
aim7 shows the following improvements in throughput:
+--------------+---------------------+-----------------+
| Workload | throughput-increase | number of users |
+--------------+---------------------+-----------------+
| alltests | 20% | >1000 |
| custom | 27%, 60% | 10-100, >1000 |
| high_systime | 36%, 30% | >100, >1000 |
| shared | 58%, 29% | 10-100, >1000 |
+--------------+---------------------+-----------------+
There was also improvement on smaller systems, such as a quad-core
x86-64 laptop running a 30Gb PostgreSQL (pgbench) workload for up
to +60% in throughput for over 50 clients. Additionally, benefits
were also noticed in exim (mail server) workloads. Furthermore, no
performance regression have been seen at all.
Based-on-work-from: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
[peterz: rej fixup due to comment patches, sched/rt.h header]
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Alex Shi <alex.shi@linaro.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Hurley <peter@hurleysoftware.com>
Cc: "Paul E.McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Jason Low <jason.low2@hp.com>
Cc: Aswin Chandramouleeswaran <aswin@hp.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: "Scott J Norton" <scott.norton@hp.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Mason <clm@fb.com>
Cc: Josef Bacik <jbacik@fusionio.com>
Link: http://lkml.kernel.org/r/1399055055.6275.15.camel@buesod1.americas.hpqcorp.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Since the smp_mb__{before,after}*() ops are fundamentally dependent on
how an arch can implement atomics it doesn't make sense to have 3
variants of them. They must all be the same.
Furthermore, the 3 variants suggest they're only valid for those 3
atomic ops, while we have many more where they could be applied.
So move away from
smp_mb__{before,after}_{atomic,clear}_{dec,inc,bit}() and reduce the
interface to just the two: smp_mb__{before,after}_atomic().
This patch prepares the way by introducing default implementations in
asm-generic/barrier.h that default to a full barrier and providing
__deprecated inlines for the previous 6 barriers if they're not
provided by the arch.
This should allow for a mostly painless transition (lots of deprecated
warns in the interim).
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/n/tip-wr59327qdyi9mbzn6x937s4e@git.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: "Chen, Gong" <gong.chen@linux.intel.com>
Cc: John Sullivan <jsrhbz@kanargh.force9.co.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mauro Carvalho Chehab <m.chehab@samsung.com>
Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: linux-arch@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull parisc updates from Helge Deller:
"There are two major changes in this patchset:
The major fix is that the epoll_pwait() syscall for 32bit userspace
was not using the compat wrapper on a 64bit kernel.
Secondly we changed the value of SHMLBA from 4MB to PAGE_SIZE to
reflect that we can actually mmap to any multiple of PAGE_SIZE. The
only thing which needs care is that shared mmaps need to be mapped at
the same offset inside the 4MB cache window"
* 'parisc-3.15' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
parisc: fix epoll_pwait syscall on compat kernel
parisc: change value of SHMLBA from 0x00400000 to PAGE_SIZE
parisc: Replace __get_cpu_var uses for address calculation
Merge ipmi fixes from Corey Minyard:
"Things collected since last kernel release.
Some of these are pretty important. The first three are bug fixes.
The next two are to hopefully make everyone happy about allowing
ACPI to be on all the time and not have IPMI have an effect on the
system when not in use. The last is a little cleanup"
* emailed patches from Corey Minyard <cminyard@mvista.com>:
ipmi: boolify some things
ipmi: Turn off all activity on an idle ipmi interface
ipmi: Turn off default probing of interfaces
ipmi: Reset the KCS timeout when starting error recovery
ipmi: Fix a race restarting the timer
Char: ipmi_bt_sm, fix infinite loop
The IPMI driver would wake up periodically looking for events and
watchdog pretimeouts. If there is nothing waiting for these events,
it's really kind of pointless to be checking for them. So modify the
driver so the message handler can pass down if it needs the lower layer
to be waiting for these. Modify the system interface lower layer to
turn off all timer and thread activity if the upper layer doesn't need
anything and it is not currently handling messages. And modify the
message handler to not restart the timer if its timer is not needed.
The timers and kthread will still be enabled if:
- the SI interface is handling a message.
- a user has enabled watching for events.
- the IPMI watchdog timer is in use (since it uses pretimeouts).
- the message handler is waiting on a remote response.
- a user has registered to receive commands.
This mostly affects interfaces without interrupts. Interfaces with
interrupts already don't use CPU in the system interface when the
interface is idle.
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The default probing can cause problems with some system, slow booting,
extra CPU usages, etc. Turn it off by default and give a config option
to enable it.
From: Matthew Garrett <matthew.garrett@nebula.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
With recent changes it is possible for the timer handler to detect an
idle interface and not start the timer, but the thread to start an
operation at the same time. The thread will not start the timer in that
instance, resulting in the timer not running.
Instead, move all timer operations under the lock and start the timer in
the thread if it detect non-idle and the timer is not already running.
Moving under locks allows the last timeout to be set in both the thread
and the timer. 'Timer is not running' means that the timer is not
pending and smi_timeout() is not running. So we need a flag to detect
this correctly.
Also fix a few other timeout bugs: setting the last timeout when the
interrupt has to be disabled and the timer started, and setting the last
timeout in check_start_timer_thread possibly racing with the timer
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Bodo Stroesser <bstroesser@ts.fujitsu.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In read_all_bytes, we do
unsigned char i;
...
bt->read_data[0] = BMC2HOST;
bt->read_count = bt->read_data[0];
...
for (i = 1; i <= bt->read_count; i++)
bt->read_data[i] = BMC2HOST;
If bt->read_data[0] == bt->read_count == 255, we loop infinitely in the
'for' loop. Make 'i' an 'int' instead of 'char' to get rid of the
overflow and finish the loop after 255 iterations every time.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Reported-and-debugged-by: Rui Hui Dian <rhdian@novell.com>
Cc: Tomas Cech <tcech@suse.cz>
Cc: Corey Minyard <minyard@acm.org>
Cc: <openipmi-developer@lists.sourceforge.net>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull Xen fixes from David Vrabel:
"Xen regression and bug fixes for 3.15-rc1:
- fix completely broken 32-bit PV guests caused by x86 refactoring
32-bit thread_info.
- only enable ticketlock slow path on Xen (not bare metal)
- fix two bugs with PV guests not shutting down when requested
- fix a minor memory leak in xen-pciback error path"
* tag 'stable/for-linus-3.15-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
xen/manage: Poweroff forcefully if user-space is not yet up.
xen/xenbus: Avoid synchronous wait on XenBus stalling shutdown/restart.
xen/spinlock: Don't enable them unconditionally.
xen-pciback: silence an unwanted debug printk
xen: fix memory leak in __xen_pcibk_add_pci_dev()
x86/xen: Fix 32-bit PV guests's usage of kernel_stack
Pull md bugfix from Neil Brown:
"One BUG fix for md for recent commit"
* tag '3.15-fixes' of git://neil.brown.name/md:
raid5: fix a race of stripe count check
Pull fbdev renaming patches from Tomi Valkeinen:
"Reorder drivers/video/ directory so that all fbdev drivers are now
located in drivers/video/fbdev/ and the fbdev framework core files are
located in drivers/video/fbdev/core/
The drivers/video/Kconfig is modified so that the DRM and the fbdev
menu options are in separate submenus, instead of both being mixed in
the same 'Graphics support' menu level"
* tag 'fbdev-reorder-3.15' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux:
video: Kconfig: move drm and fb into separate menus
fbdev: move fbdev core files to separate directory
video: move fbdev to drivers/video/fbdev
I hit another BUG_ON with e240c1839d. In __get_priority_stripe(),
stripe count equals to 0 initially. Between atomic_inc and BUG_ON,
get_active_stripe() finds the stripe. So the stripe count isn't 1 any more.
V2: keeps the BUG_ON suggested by Neil.
Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: NeilBrown <neilb@suse.de>
At the moment the "Device Drivers / Graphics support" kernel config page
looks rather messy, with DRM and fbdev driver selections on the same
page, some on the top level Graphics support page, some under their
respective subsystems.
If I'm not mistaken, this is caused by the drivers depending on other
things than DRM or FB, which causes Kconfig to arrange the options in
not-so-neat manner.
Both DRM and FB have a main menuconfig option for the whole DRM or FB
subsystem. Optimally, this would be enough to arrange all DRM and FB
options under the respective subsystem, but for whatever reason this
doesn't work reliably.
This patch adds an explicit submenu for DRM and FB, making it much
clearer which options are related to FB, and which to DRM.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Jingoo Han <jg1.han@samsung.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Instead of having fbdev framework core files at the root fbdev
directory, mixed with random fbdev device drivers, move the fbdev core
files to a separate core directory. This makes it much clearer which of
the files are actually part of the fbdev framework, and which are part
of device drivers.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Rob Clark <robdclark@gmail.com>
Acked-by: Jingoo Han <jg1.han@samsung.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
