In the case of x2apic cluster mode we can group IPI register
writes based on the cluster group instead of individual per-cpu
destination messages.
This reduces the apic register writes and reduces the amount of
IPI messages (in the best case we can reduce it by a factor of
16).
With this change, the cost of flush_tlb_others(), with the flush
tlb IPI being sent from a cpu in the socket-1 to all the logical
cpus in socket-2 (on a Westmere-EX system that has 20 logical
cpus in a socket) is 3x times better now (compared to the former
'send one-by-one' algorithm).
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: steiner@sgi.com
Cc: yinghai@kernel.org
Link: http://lkml.kernel.org/r/20110519234637.512271057@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In the case of x2apic cluster mode, we can group IPI register
writes based on the cluster group instead of individual per-cpu
destination messages.
For this purpose, track the cpu's that belong to the same x2apic
cluster.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: steiner@sgi.com
Cc: yinghai@kernel.org
Link: http://lkml.kernel.org/r/20110519234637.421800999@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Use the unused probe routine in the apic driver to finalize the
apic model selection. This cleans up the
default_setup_apic_routing() and this probe routine in future
can also be used for doing any apic model specific
initialisation.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: steiner@sgi.com
Cc: yinghai@kernel.org
Link: http://lkml.kernel.org/r/20110519234637.247458931@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Code flow for enabling interrupt-remapping has its own routines
for saving and restoring io-apic RTE's. ioapic suspend/resume
code flow also has similar routines. Remove the duplicate code.
Tested-by: Daniel J Blueman <daniel.blueman@gmail.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/20110518233157.673130611@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Code flow for enabling interrupt-remapping was
allocating/freeing buffers for saving/restoring io-apic RTE's.
ioapic suspend/resume code uses boot time allocated
ioapic_saved_data that is a perfect match for reuse here.
This will remove the unnecessary allocation/free of the
temporary buffers during suspend/resume of interrupt-remapping
enabled platforms aswell as paving the way for further code
consolidation.
Tested-by: Daniel J Blueman <daniel.blueman@gmail.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/20110518233157.574469296@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This allows re-using this buffer for enabling
interrupt-remapping during boot and resume. And thus allow for
consolidating the code between ioapic suspend/resume and
interrupt-remapping.
Tested-by: Daniel J Blueman <daniel.blueman@gmail.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/20110518233157.481404505@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fix a potential deadlock when resuming; here the calling
function has disabled interrupts, so we cannot sleep.
Change the memory allocation flag from GFP_KERNEL to GFP_ATOMIC.
TODO: We can do away with this memory allocation during resume
by reusing the ioapic suspend/resume code that uses boot time
allocated buffers, but we want to keep this -stable patch
simple.
Signed-off-by: Daniel J Blueman <daniel.blueman@gmail.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: <stable@kernel.org> # v2.6.38/39
Link: http://lkml.kernel.org/r/20110518233157.385970138@sbsiddha-MOBL3.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
There is no point in counting hits - we can calculate it from the number
of lookups and misses.
This makes the avc statistics a bit smaller, and makes the code
generation better too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
You can turn off the avc cache stats, but distributions seem to not do
that (perhaps because several performance tuning how-to's talk about the
avc cache statistics).
Which is sad, because the code it generates is truly horrendous, with
the statistics update being sandwitched between get_cpu/put_cpu which in
turn causes preemption disables etc. We're talking ten+ instructions
just to increment a per-cpu variable in some pretty hot code.
Fix the craziness by just using 'this_cpu_inc()' instead. Suddenly we
only need a single 'inc' instruction to increment the statistics. This
is quite noticeable in the incredibly hot avc_has_perm_noaudit()
function (which triggers all the statistics by virtue of doing an
avc_lookup() call).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'driver-core-next' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core-2.6: (44 commits)
debugfs: Silence DEBUG_STRICT_USER_COPY_CHECKS=y warning
sysfs: remove "last sysfs file:" line from the oops messages
drivers/base/memory.c: fix warning due to "memory hotplug: Speed up add/remove when blocks are larger than PAGES_PER_SECTION"
memory hotplug: Speed up add/remove when blocks are larger than PAGES_PER_SECTION
SYSFS: Fix erroneous comments for sysfs_update_group().
driver core: remove the driver-model structures from the documentation
driver core: Add the device driver-model structures to kerneldoc
Translated Documentation/email-clients.txt
RAW driver: Remove call to kobject_put().
reboot: disable usermodehelper to prevent fs access
efivars: prevent oops on unload when efi is not enabled
Allow setting of number of raw devices as a module parameter
Introduce CONFIG_GOOGLE_FIRMWARE
driver: Google Memory Console
driver: Google EFI SMI
x86: Better comments for get_bios_ebda()
x86: get_bios_ebda_length()
misc: fix ti-st build issues
params.c: Use new strtobool function to process boolean inputs
debugfs: move to new strtobool
...
Fix up trivial conflicts in fs/debugfs/file.c due to the same patch
being applied twice, and an unrelated cleanup nearby.
This fixes these build errors on powerpc:
In file included from arch/powerpc/mm/fault.c:18:
include/linux/signal.h:239: error: 'struct task_struct' declared inside parameter list
include/linux/signal.h:239: error: its scope is only this definition or declaration, which is probably not what you want
include/linux/signal.h:240: error: 'struct task_struct' declared inside parameter list
..
Exposed by commit e66eed651f ("list: remove prefetching from regular
list iterators"), which removed the include of <linux/prefetch.h> from
<linux/list.h>.
Without that, linux/signal.h no longer accidentally got the declaration
of 'struct task_struct'.
Fix by properly declaring the struct, rather than introducing any new
header file dependency.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'x86-platform-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: Introduce pci_map_biosrom()
x86, olpc: Use device tree for platform identification
* 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (50 commits)
x86, mm: Allow ZONE_DMA to be configurable
x86, NUMA: Trim numa meminfo with max_pfn in a separate loop
x86, NUMA: Rename setup_node_bootmem() to setup_node_data()
x86, NUMA: Enable emulation on 32bit too
x86, NUMA: Enable CONFIG_AMD_NUMA on 32bit too
x86, NUMA: Rename amdtopology_64.c to amdtopology.c
x86, NUMA: Make numa_init_array() static
x86, NUMA: Make 32bit use common NUMA init path
x86, NUMA: Initialize and use remap allocator from setup_node_bootmem()
x86-32, NUMA: Add @start and @end to init_alloc_remap()
x86, NUMA: Remove long 64bit assumption from numa.c
x86, NUMA: Enable build of generic NUMA init code on 32bit
x86, NUMA: Move NUMA init logic from numa_64.c to numa.c
x86-32, NUMA: Update numaq to use new NUMA init protocol
x86-32, NUMA: Replace srat_32.c with srat.c
x86-32, NUMA: implement temporary NUMA init shims
x86, NUMA: Move numa_nodes_parsed to numa.[hc]
x86-32, NUMA: Move get_memcfg_numa() into numa_32.c
x86, NUMA: make srat.c 32bit safe
x86, NUMA: rename srat_64.c to srat.c
...
* 'x86-efi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, efi: Ensure that the entirity of a region is mapped
x86, efi: Pass a minimal map to SetVirtualAddressMap()
x86, efi: Merge contiguous memory regions of the same type and attribute
x86, efi: Consolidate EFI nx control
x86, efi: Remove virtual-mode SetVirtualAddressMap call
* 'x86-gart-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, gart: Don't enforce GART aperture lower-bound by alignment
* 'x86-irq-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: Don't unmask disabled irqs when migrating them
x86: Skip migrating IRQF_PER_CPU irqs in fixup_irqs()
* 'x86-mce-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, mce: Drop the default decoding notifier
x86, MCE: Do not taint when handling correctable errors
* 'x86-cpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, cpu: Fix detection of Celeron Covington stepping A1 and B0
Documentation, ABI: Update L3 cache index disable text
x86, AMD, cacheinfo: Fix L3 cache index disable checks
x86, AMD, cacheinfo: Fix fallout caused by max3 conversion
x86, cpu: Change NOP selection for certain Intel CPUs
x86, cpu: Clean up and unify the NOP selection infrastructure
x86, percpu: Use ASM_NOP4 instead of hardcoding P6_NOP4
x86, cpu: Move AMD Elan Kconfig under "Processor family"
Fix up trivial conflicts in alternative handling (commit dc326fca2b
"x86, cpu: Clean up and unify the NOP selection infrastructure" removed
some hacky 5-byte instruction stuff, while commit d430d3d7e6 "jump
label: Introduce static_branch() interface" renamed HAVE_JUMP_LABEL to
CONFIG_JUMP_LABEL in the code that went away)
* 'timers-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
hrtimer: Make lookup table const
RTC: Disable CONFIG_RTC_CLASS from being built as a module
timers: Fix alarmtimer build issues when CONFIG_RTC_CLASS=n
timers: Remove delayed irqwork from alarmtimers implementation
timers: Improve alarmtimer comments and minor fixes
timers: Posix interface for alarm-timers
timers: Introduce in-kernel alarm-timer interface
timers: Add rb_init_node() to allow for stack allocated rb nodes
time: Add timekeeping_inject_sleeptime
* 'timers-clockevents-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: hpet: Cleanup the clockevents init and register code
x86: Convert PIT to clockevents_config_and_register()
clockevents: Provide interface to reconfigure an active clock event device
clockevents: Provide combined configure and register function
clockevents: Restructure clock_event_device members
clocksource: Get rid of the hardcoded 5 seconds sleep time limit
clocksource: Restructure clocksource struct members
* 'timers-clocksource-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
clocksource: convert mips to generic i8253 clocksource
clocksource: convert x86 to generic i8253 clocksource
clocksource: convert footbridge to generic i8253 clocksource
clocksource: add common i8253 PIT clocksource
blackfin: convert to clocksource_register_hz
mips: convert to clocksource_register_hz/khz
sparc: convert to clocksource_register_hz/khz
alpha: convert to clocksource_register_hz
microblaze: convert to clocksource_register_hz/khz
ia64: convert to clocksource_register_hz/khz
x86: Convert remaining x86 clocksources to clocksource_register_hz/khz
Make clocksource name const
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (60 commits)
sched: Fix and optimise calculation of the weight-inverse
sched: Avoid going ahead if ->cpus_allowed is not changed
sched, rt: Update rq clock when unthrottling of an otherwise idle CPU
sched: Remove unused parameters from sched_fork() and wake_up_new_task()
sched: Shorten the construction of the span cpu mask of sched domain
sched: Wrap the 'cfs_rq->nr_spread_over' field with CONFIG_SCHED_DEBUG
sched: Remove unused 'this_best_prio arg' from balance_tasks()
sched: Remove noop in alloc_rt_sched_group()
sched: Get rid of lock_depth
sched: Remove obsolete comment from scheduler_tick()
sched: Fix sched_domain iterations vs. RCU
sched: Next buddy hint on sleep and preempt path
sched: Make set_*_buddy() work on non-task entities
sched: Remove need_migrate_task()
sched: Move the second half of ttwu() to the remote cpu
sched: Restructure ttwu() some more
sched: Rename ttwu_post_activation() to ttwu_do_wakeup()
sched: Remove rq argument from ttwu_stat()
sched: Remove rq->lock from the first half of ttwu()
sched: Drop rq->lock from sched_exec()
...
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: Fix rt_rq runtime leakage bug
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (107 commits)
perf stat: Add more cache-miss percentage printouts
perf stat: Add -d -d and -d -d -d options to show more CPU events
ftrace/kbuild: Add recordmcount files to force full build
ftrace: Add self-tests for multiple function trace users
ftrace: Modify ftrace_set_filter/notrace to take ops
ftrace: Allow dynamically allocated function tracers
ftrace: Implement separate user function filtering
ftrace: Free hash with call_rcu_sched()
ftrace: Have global_ops store the functions that are to be traced
ftrace: Add ops parameter to ftrace_startup/shutdown functions
ftrace: Add enabled_functions file
ftrace: Use counters to enable functions to trace
ftrace: Separate hash allocation and assignment
ftrace: Create a global_ops to hold the filter and notrace hashes
ftrace: Use hash instead for FTRACE_FL_FILTER
ftrace: Replace FTRACE_FL_NOTRACE flag with a hash of ignored functions
perf bench, x86: Add alternatives-asm.h wrapper
x86, 64-bit: Fix copy_[to/from]_user() checks for the userspace address limit
x86, mem: memset_64.S: Optimize memset by enhanced REP MOVSB/STOSB
x86, mem: memmove_64.S: Optimize memmove by enhanced REP MOVSB/STOSB
...
* 'core-locking-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
seqlock: Don't smp_rmb in seqlock reader spin loop
watchdog, hung_task_timeout: Add Kconfig configurable default
lockdep: Remove cmpxchg to update nr_chain_hlocks
lockdep: Print a nicer description for simple irq lock inversions
lockdep: Replace "Bad BFS generated tree" message with something less cryptic
lockdep: Print a nicer description for irq inversion bugs
lockdep: Print a nicer description for simple deadlocks
lockdep: Print a nicer description for normal deadlocks
lockdep: Print a nicer description for irq lock inversions
* 'core-iommu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, gart: Rename pci-gart_64.c to amd_gart_64.c
x86/amd-iommu: Use threaded interupt handler
arch/x86/kernel/pci-iommu_table.c: Convert sprintf_symbol to %pS
x86/amd-iommu: Add support for invalidate_all command
x86/amd-iommu: Add extended feature detection
x86/amd-iommu: Add ATS enable/disable code
x86/amd-iommu: Add flag to indicate IOTLB support
x86/amd-iommu: Flush device IOTLB if ATS is enabled
x86/amd-iommu: Select PCI_IOV with AMD IOMMU driver
PCI: Move ATS declarations in seperate header file
dma-debug: print information about leaked entry
x86/amd-iommu: Flush all internal TLBs when IOMMUs are enabled
x86/amd-iommu: Rename iommu_flush_device
x86/amd-iommu: Improve handling of full command buffer
x86/amd-iommu: Rename iommu_flush* to domain_flush*
x86/amd-iommu: Remove command buffer resetting logic
x86/amd-iommu: Cleanup completion-wait handling
x86/amd-iommu: Cleanup inv_pages command handling
x86/amd-iommu: Move inv-dte command building to own function
x86/amd-iommu: Move compl-wait command building to own function
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/suspend-2.6: (34 commits)
PM: Introduce generic prepare and complete callbacks for subsystems
PM: Allow drivers to allocate memory from .prepare() callbacks safely
PM: Remove CONFIG_PM_VERBOSE
Revert "PM / Hibernate: Reduce autotuned default image size"
PM / Hibernate: Add sysfs knob to control size of memory for drivers
PM / Wakeup: Remove useless synchronize_rcu() call
kmod: always provide usermodehelper_disable()
PM / ACPI: Remove acpi_sleep=s4_nonvs
PM / Wakeup: Fix build warning related to the "wakeup" sysfs file
PM: Print a warning if firmware is requested when tasks are frozen
PM / Runtime: Rework runtime PM handling during driver removal
Freezer: Use SMP barriers
PM / Suspend: Do not ignore error codes returned by suspend_enter()
PM: Fix build issue in clock_ops.c for CONFIG_PM_RUNTIME unset
PM: Revert "driver core: platform_bus: allow runtime override of dev_pm_ops"
OMAP1 / PM: Use generic clock manipulation routines for runtime PM
PM: Remove sysdev suspend, resume and shutdown operations
PM / PowerPC: Use struct syscore_ops instead of sysdevs for PM
PM / UNICORE32: Use struct syscore_ops instead of sysdevs for PM
PM / AVR32: Use struct syscore_ops instead of sysdevs for PM
...
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
IB/qib: Use pci_dev->revision
RDMA/iwcm: Get rid of enum iw_cm_event_status
IB/ipath: Use pci_dev->revision, again
IB/qib: Prevent driver hang with unprogrammed boards
RDMA/cxgb4: EEH errors can hang the driver
RDMA/cxgb4: Reset wait condition atomically
RDMA/cxgb4: Fix missing parentheses
RDMA/cxgb4: Initialization errors can cause crash
RDMA/cxgb4: Don't change QP state outside EP lock
RDMA/cma: Add an ID_REUSEADDR option
RDMA/cma: Fix handling of IPv6 addressing in cma_use_port
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cmarinas/linux-2.6-cm:
kmemleak: Initialise kmemleak after debug_objects_mem_init()
kmemleak: Select DEBUG_FS unconditionally in DEBUG_KMEMLEAK
kmemleak: Do not return a pointer to an object that kmemleak did not get
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k:
input/atari: Fix mouse movement and button mapping
input/atari: Fix atarimouse init
input/atari: Use the correct mouse interrupt hook
m68k/atari: Do not use "/" in interrupt names
m68k: unistd - Comment out definitions for unimplemented syscalls
m68k: Really wire up sys_pselect6 and sys_ppoll
m68k: Merge mmu and non-mmu versions of sys_call_table
MAINTAINERS: Roman Zippel has been MIA for several years.
m68k: bitops - Never step beyond the end of the bitmap
m68k: bitops - offset == ((long)p - (long)vaddr) * 8
* 'stable/irq' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen: do not clear and mask evtchns in __xen_evtchn_do_upcall
* 'stable/p2m.bugfixes' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/p2m: Create entries in the P2M_MFN trees's to track 1-1 mappings
* 'stable/e820.bugfixes' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/setup: Fix for incorrect xen_extra_mem_start initialization under 32-bit
xen/setup: Ignore E820_UNUSABLE when setting 1-1 mappings.
* 'stable/mmu.bugfixes' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen mmu: fix a race window causing leave_mm BUG()
* 'stable/backend.base.v3' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/pci: Fix compiler error when CONFIG_XEN_PRIVILEGED_GUEST is not set.
xen/p2m: Add EXPORT_SYMBOL_GPL to the M2P override functions.
xen/p2m/m2p/gnttab: Support GNTMAP_host_map in the M2P override.
xen/irq: The Xen hypervisor cleans up the PIRQs if the other domain forgot.
xen/irq: Export 'xen_pirq_from_irq' function.
xen/irq: Add support to check if IRQ line is shared with other domains.
xen/irq: Check if the PCI device is owned by a domain different than DOMID_SELF.
xen/pci: Add xen_[find|register|unregister]_device_domain_owner functions.
* 'stable/gntalloc.v7' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/gntdev,gntalloc: Remove unneeded VM flags
* git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus:
params.c: Use new strtobool function to process boolean inputs
debugfs: move to new strtobool
Add a strtobool function matching semantics of existing in kernel equivalents
modpost: Update 64k section support for binutils 2.18.50
module: Use binary search in lookup_symbol()
module: Use the binary search for symbols resolution
lib: Add generic binary search function to the kernel.
module: Sort exported symbols
module: each_symbol_section instead of each_symbol
module: split unset_section_ro_nx function.
module: undo module RONX protection correctly.
module: zero mod->init_ro_size after init is freed.
minor ANSI prototype sparse fix
module: reorder kparam_array to remove alignment padding on 64 bit builds
module: remove 64 bit alignment padding from struct module with CONFIG_TRACE*
module: do not hide __modver_version_show declaration behind ifdef
module: deal with alignment issues in built-in module versions
* 'docs-move' of git://git.kernel.org/pub/scm/linux/kernel/git/rdunlap/linux-docs:
Correct occurrences of - Documentation/kvm/ to Documentation/virtual/kvm - Documentation/uml/ to Documentation/virtual/uml - Documentation/lguest/ to Documentation/virtual/lguest throughout the kernel source tree.
Add a 00-INDEX file to Documentation/virtual Remove uml from the top level 00-INDEX file.
Move kvm, uml, and lguest subdirectories under a common "virtual" directory, I.E:
This reverts commit e59fb3120b.
This reversion was due to (extreme) boot-time slowdowns on SPARC seen by
Yinghai Lu and on x86 by Ingo
.
This is a non-trivial reversion due to intervening commits.
Conflicts:
Documentation/RCU/trace.txt
kernel/rcutree.c
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This is removes the use of software prefetching from the regular list
iterators. We don't want it. If you do want to prefetch in some
iterator of yours, go right ahead. Just don't expect the iterator to do
it, since normally the downsides are bigger than the upsides.
It also replaces <linux/prefetch.h> with <linux/const.h>, because the
use of LIST_POISON ends up needing it. <linux/poison.h> is sadly not
self-contained, and including prefetch.h just happened to hide that.
Suggested by David Miller (networking has a lot of regular lists that
are often empty or a single entry, and prefetching is not going to do
anything but add useless instructions).
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: linux-arch@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
They not only increase the code footprint, they actually make things
slower rather than faster. On internationally acclaimed benchmarks
("make -j16" on an already fully built kernel source tree) the hlist
prefetching slows down the build by up to 1%.
(Almost all of it comes from hlist_for_each_entry_rcu() as used by
avc_has_perm_noaudit(), which is very hot due to all the pathname
lookups to see if there is anything to do).
The cause seems to be two-fold:
- on at least some Intel cores, prefetch(NULL) ends up with some
microarchitectural stall due to the TLB miss that it incurs. The
hlist case triggers this very commonly, since the NULL pointer is the
last entry in the list.
- the prefetch appears to cause more D$ activity, probably because it
prefetches hash list entries that are never actually used (because we
ended the search early due to a hit).
Regardless, the numbers clearly say that the implicit prefetching is
simply a bad idea. If some _particular_ user of the hlist iterators
wants to prefetch the next list entry, they can do so themselves
explicitly, rather than depend on all list iterators doing so
implicitly.
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: linux-arch@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
