There was some desire in large applications using MAP_HUGETLB or
SHM_HUGETLB to use 1GB huge pages on some mappings, and stay with 2MB on
others. This is useful together with NUMA policy: use 2MB interleaving
on some mappings, but 1GB on local mappings.
This patch extends the IPC/SHM syscall interfaces slightly to allow
specifying the page size.
It borrows some upper bits in the existing flag arguments and allows
encoding the log of the desired page size in addition to the *_HUGETLB
flag. When 0 is specified the default size is used, this makes the
change fully compatible.
Extending the internal hugetlb code to handle this is straight forward.
Instead of a single mount it just keeps an array of them and selects the
right mount based on the specified page size. When no page size is
specified it uses the mount of the default page size.
The change is not visible in /proc/mounts because internal mounts don't
appear there. It also has very little overhead: the additional mounts
just consume a super block, but not more memory when not used.
I also exported the new flags to the user headers (they were previously
under __KERNEL__). Right now only symbols for x86 and some other
architecture for 1GB and 2MB are defined. The interface should already
work for all other architectures though. Only architectures that define
multiple hugetlb sizes actually need it (that is currently x86, tile,
powerpc). However tile and powerpc have user configurable hugetlb
sizes, so it's not easy to add defines. A program on those
architectures would need to query sysfs and use the appropiate log2.
[[email protected]: cleanups]
[[email protected]: fix build]
[[email protected]: checkpatch fixes]
Signed-off-by: Andi Kleen <[email protected]>
Cc: Michael Kerrisk <[email protected]>
Acked-by: Rik van Riel <[email protected]>
Acked-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Hillf Danton <[email protected]>
Signed-off-by: David Rientjes <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The PA-RISC tool chain seems to have some problem with correct
read/write attributes on sections. This causes problems when the const
sections are fixed up for other architecture to only contain truly
read-only data.
Disable const sections for PA-RISC
This can cause a bit of noise with modpost.
Signed-off-by: Andi Kleen <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Cc: Helge Deller <[email protected]>
Acked-by: Sam Ravnborg <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
For large kernel configurations (like a distribution kernel)
targz-pkg takes a quite long time to just do the compression.
I clocked it at 15+mins for a SUSE kernel like config on a fast
system. And tarxz and bzip2 are even slower.
The main reason is that the script that is doing the taring sets
the highest compression level (-9). When I change it to just
use the defaults the gzip time for the same kernel goes down
to ~3 mins. I haven't tested xz and bzip, but I expect those
to be much faster too.
I'm not willing to wait that long for a small compression
gain. So just change the script to use the defaults.
Signed-off-by: Andi Kleen <[email protected]>
Reviewed-by: Fengguang Wu <[email protected]>
Signed-off-by: Michal Marek <[email protected]>
gcc 4.6+ has support for a externally_visible attribute that prevents the
optimizer from optimizing unused symbols away. Add a __visible macro to
use it with that compiler version or later.
This is used (at least) by the "Link Time Optimization" patchset.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The new LTO EXPORT_SYMBOL references symbols even without CONFIG_MODULES.
Since these functions are macros in this case this doesn't work.
Add a ifdef to fix the build.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Takashi Iwai <[email protected]>
floppy_init is quite slow, 3s on my test system to determine
that there is no floppy. Run it asynchronous to the other
init calls to improve boot time.
[[email protected]: fix modular build]
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Jiri Kosina <[email protected]>
slab_node() could access current->mempolicy from interrupt context.
However there's a race condition during exit where the mempolicy
is first freed and then the pointer zeroed.
Using this from interrupts seems bogus anyways. The interrupt
will interrupt a random process and therefore get a random
mempolicy. Many times, this will be idle's, which noone can change.
Just disable this here and always use local for slab
from interrupts. I also cleaned up the callers of slab_node a bit
which always passed the same argument.
I believe the original mempolicy code did that in fact,
so it's likely a regression.
v2: send version with correct logic
v3: simplify. fix typo.
Reported-by: Arun Sharma <[email protected]>
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Andi Kleen <[email protected]>
[[email protected]: Rework control flow based on feedback from
[email protected], fix logic, and cleanup current task_struct reference]
Acked-by: David Rientjes <[email protected]>
Acked-by: Christoph Lameter <[email protected]>
Acked-by: KOSAKI Motohiro <[email protected]>
Signed-off-by: David Mackey <[email protected]>
Signed-off-by: Pekka Enberg <[email protected]>
On Sandy Bridge in non HT mode there are 8 counters available.
Since every counter can write a PEBS record assuming there are
4 max is incorrect. Use the reported counter number -- with an
upper limit for a static array -- instead.
Also I made the warning messages a bit more informational.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
lglocks and brlocks are currently generated with some complicated macros
in lglock.h. But there's no reason to not just use common utility
functions and put all the data into a common data structure.
In preparation, this patch changes the API to look more like normal
function calls with pointers, not magic macros.
The patch is rather large because I move over all users in one go to keep
it bisectable. This impacts the VFS somewhat in terms of lines changed.
But no actual behaviour change.
[[email protected]: checkpatch fixes]
Signed-off-by: Andi Kleen <[email protected]>
Cc: Al Viro <[email protected]>
Cc: Rusty Russell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Rusty Russell <[email protected]>
Signed-off-by: Al Viro <[email protected]>
lglocks and brlocks are currently generated with some complicated macros
in lglock.h. But there's no reason to not just use common utility
functions and put all the data into a common data structure.
Since there are at least two users it makes sense to share this code in a
library. This is also easier maintainable than a macro forest.
This will also make it later possible to dynamically allocate lglocks and
also use them in modules (this would both still need some additional, but
now straightforward, code)
[[email protected]: checkpatch fixes]
Signed-off-by: Andi Kleen <[email protected]>
Cc: Al Viro <[email protected]>
Cc: Rusty Russell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Rusty Russell <[email protected]>
Signed-off-by: Al Viro <[email protected]>
When ata_generic_ide=1 is set don't do the is_intel_ider() magic
check. We found at least one box who needed that.
Cc: [email protected]
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Jeff Garzik <[email protected]>
On a system on the thermal limit these are quite noisy and flood the logs.
Better would be a counter anyways. But given that we don't even have
anything for normal throttling this doesn't seem to be urgent either.
Reviewed-by: Jesse Barnes <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Len Brown <[email protected]>
Add a config option to disable various gcc compiler optimizations that
make assembler listings much harder to read. This is everything that reorders
code significantly or creates partial functions.
This is mainly to keep kernel hackers sane.
Signed-off-by: Andi Kleen <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: H. Peter Anvin <[email protected]>
On gccs that support AVX it's a good idea to disable that too, similar to
how SSE2, SSE1 etc. are already disabled. This prevents the compiler
from generating AVX ever implicitely.
No failure observed, just from review.
[ hpa: Marking this for urgent and stable, simply because the patch
will either have absolutely no effect *or* it will avoid potentially
very hard to debug failures. ]
Signed-off-by: Andi Kleen <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: H. Peter Anvin <[email protected]>
Cc: <[email protected]>
When i_mmap_lock changed to a mutex the locking order in memory failure
was changed to take the sleeping lock first. But the big fat mm lock
ordering comment (BFMLO) wasn't updated. Do this here.
Pointed out by Andrew.
Signed-off-by: Andi Kleen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This marks all the x86 cpuinfo tables to the CPU specific device drivers,
to allow auto loading by udev. This should simplify the distribution
startup scripts for this greatly.
I didn't add MODULE_DEVICE_IDs to the centrino and p4-clockmod drivers,
because those probably shouldn't be auto loaded and the acpi driver
be used instead (not fully sure on that, would appreciate feedback)
The old nforce drivers autoload based on the PCI ID.
ACPI cpufreq is autoloaded in another patch.
v3: Autoload gx based on PCI IDs only. Remove cpu check (Dave Jones)
v4: Use newly introduce HW_PSTATE feature for powernow-k8 loading
Cc: Dave Jones <[email protected]>
Cc: Kay Sievers <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Thomas Renninger <[email protected]>
Acked-by: H. Peter Anvin <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
The only left over hole in automatic cpufreq driver loading was the loading
of ACPI cpufreq. This driver should be loaded when ACPI supports a _PDC
method and the CPU vendor wants to use acpi cpufreq.
Simply add a request module call to the acpi processor core driver
when this is true. This seems like the simplest solution for this.
Cc: Len Brown <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Thomas Renninger <[email protected]>
Acked-by: H. Peter Anvin <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
With this it should be automatically loaded on suitable systems by
udev.
The old switch () is replaced with a table based approach, this
also cleans up the code.
Cc: Len Brown <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Thomas Renninger <[email protected]>
Acked-by: H. Peter Anvin <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Add support for auto-loading of crypto drivers based on cpuid features.
This enables auto-loading of the VIA and Intel specific drivers
for AES, hashing and CRCs.
Requires the earlier infrastructure patch to add x86 modinfo.
I kept it all in a single patch for now.
I dropped the printks when the driver cpuid doesn't match (imho
drivers never should print anything in such a case)
One drawback is that udev doesn't know if the drivers are used or not,
so they will be unconditionally loaded at boot up. That's better
than not loading them at all, like it often happens.
Cc: Dave Jones <[email protected]>
Cc: Kay Sievers <[email protected]>
Cc: Jen Axboe <[email protected]>
Cc: Herbert Xu <[email protected]>
Cc: Huang Ying <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Thomas Renninger <[email protected]>
Acked-by: H. Peter Anvin <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
There's a growing number of drivers that support a specific x86 feature
or CPU. Currently loading these drivers currently on a generic
distribution requires various driver specific hacks and it often
doesn't work.
This patch adds auto probing for drivers based on the x86 cpuid
information, in particular based on vendor/family/model number
and also based on CPUID feature bits.
For example a common issue is not loading the SSE 4.2 accelerated
CRC module: this can significantly lower the performance of BTRFS
which relies on fast CRC.
Another issue is loading the right CPUFREQ driver for the current CPU.
Currently distributions often try all all possible driver until
one sticks, which is not really a good way to do this.
It works with existing udev without any changes. The code
exports the x86 information as a generic string in sysfs
that can be matched by udev's pattern matching.
This scheme does not support numeric ranges, so if you want to
handle e.g. ranges of model numbers they have to be encoded
in ASCII or simply all models or families listed. Fixing
that would require changing udev.
Another issue is that udev will happily load all drivers that match,
there is currently no nice way to stop a specific driver from
being loaded if it's not needed (e.g. if you don't need fast CRC)
But there are not that many cpu specific drivers around and they're
all not that bloated, so this isn't a particularly serious issue.
Originally this patch added the modalias to the normal cpu
sysdevs. However sysdevs don't have all the infrastructure
needed for udev, so it couldn't really autoload drivers.
This patch instead adds the CPU modaliases to the cpuid devices,
which are real devices with full support for udev. This implies
that the cpuid driver has to be loaded to use this.
This patch just adds infrastructure, some driver conversions
in followups.
Thanks to Kay for helping with some sysfs magic.
v2: Constifcation, some updates
v4: ([email protected]):
- Use kzalloc instead of kmalloc to terminate modalias buffer
- Use uppercase hex values to match correctly against hex values containing
letters
Cc: Dave Jones <[email protected]>
Cc: Kay Sievers <[email protected]>
Cc: Jen Axboe <[email protected]>
Cc: Herbert Xu <[email protected]>
Cc: Huang Ying <[email protected]>
Cc: Len Brown <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Thomas Renninger <[email protected]>
Acked-by: H. Peter Anvin <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Some investigation of a transaction processing workload showed that a
major consumer of cycles in __blockdev_direct_IO is the cache miss while
accessing the block size. This is because it has to walk the chain from
block_dev to gendisk to queue.
The block size is needed early on to check alignment and sizes. It's only
done if the check for the inode block size fails. But the costly block
device state is unconditionally fetched.
- Reorganize the code to only fetch block dev state when actually
needed.
Then do a prefetch on the block dev early on in the direct IO path. This
is worth it, because there is substantial code run before we actually
touch the block dev now.
- I also added some unlikelies to make it clear the compiler that block
device fetch code is not normally executed.
This gave a small, but measurable improvement on a large database
benchmark (about 0.3%)
[[email protected]: coding-style fixes]
[[email protected]: using prefetch requires including prefetch.h]
Signed-off-by: Andi Kleen <[email protected]>
Cc: Jeff Moyer <[email protected]>
Cc: Jens Axboe <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Signed-off-by: Stephen Rothwell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This makes it possible to get from the inode to the request_queue with one
less cache miss. Used in followon optimization.
The livetime of the pointer is the same as the gendisk.
This assumes that the queue will always stay the same in the gendisk while
it's visible to block_devices. I think that's safe correct?
Signed-off-by: Andi Kleen <[email protected]>
Acked-by: Jeff Moyer <[email protected]>
Cc: Jens Axboe <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
When an oops causes a panic and panic prints another backtrace it's pretty
common to have the original oops data be scrolled away on a 80x50 screen.
The second backtrace is quite redundant and not needed anyways.
So don't print the panic backtrace when oops_in_progress is true.
[[email protected]: add comment]
Signed-off-by: Andi Kleen <[email protected]>
Cc: Michael Holzheu <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Add a generic_file_llseek variant to the VFS that allows passing in
the maximum file size of the file system, instead of always
using maxbytes from the superblock.
This can be used to eliminate some cut'n'paste seek code in ext4.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
The i_mutex lock use of generic _file_llseek hurts. Independent processes
accessing the same file synchronize over a single lock, even though
they have no need for synchronization at all.
Under high utilization this can cause llseek to scale very poorly on larger
systems.
This patch does some rethinking of the llseek locking model:
First the 64bit f_pos is not necessarily atomic without locks
on 32bit systems. This can already cause races with read() today.
This was discussed on linux-kernel in the past and deemed acceptable.
The patch does not change that.
Let's look at the different seek variants:
SEEK_SET: Doesn't really need any locking.
If there's a race one writer wins, the other loses.
For 32bit the non atomic update races against read()
stay the same. Without a lock they can also happen
against write() now. The read() race was deemed
acceptable in past discussions, and I think if it's
ok for read it's ok for write too.
=> Don't need a lock.
SEEK_END: This behaves like SEEK_SET plus it reads
the maximum size too. Reading the maximum size would have the
32bit atomic problem. But luckily we already have a way to read
the maximum size without locking (i_size_read), so we
can just use that instead.
Without i_mutex there is no synchronization with write() anymore,
however since the write() update is atomic on 64bit it just behaves
like another racy SEEK_SET. On non atomic 32bit it's the same
as SEEK_SET.
=> Don't need a lock, but need to use i_size_read()
SEEK_CUR: This has a read-modify-write race window
on the same file. One could argue that any application
doing unsynchronized seeks on the same file is already broken.
But for the sake of not adding a regression here I'm
using the file->f_lock to synchronize this. Using this
lock is much better than the inode mutex because it doesn't
synchronize between processes.
=> So still need a lock, but can use a f_lock.
This patch implements this new scheme in generic_file_llseek.
I dropped generic_file_llseek_unlocked and changed all callers.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
This doesn't change anything for the compiler, but hch thought it would
make the code clearer.
I moved the reference counting into its own little inline.
Signed-off-by: Andi Kleen <[email protected]>
Acked-by: Jeff Moyer <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
Add inlines to all the submission path functions. While this increases
code size it also gives gcc a lot of optimization opportunities
in this critical hotpath.
In particular -- together with some other changes -- this
allows gcc to get rid of the unnecessary clearing of
sdio at the beginning and optimize the messy parameter passing.
Any non inlining of a function which takes a sdio parameter
would break this optimization because they cannot be done if the
address of a structure is taken.
Note that benefits are only seen with CONFIG_OPTIMIZE_INLINING
and CONFIG_CC_OPTIMIZE_FOR_SIZE both set to off.
This gives about 2.2% improvement on a large database benchmark
with a high IOPS rate.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
Only a single b_private field in the map_bh buffer head is needed after
the submission path. Move map_bh separately to avoid storing
this information in the long term slab.
This avoids the weird 104 byte hole in struct dio_submit which also needed
to be memseted early.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
A direct slab call is slightly faster than kmalloc and can be better cached
per CPU. It also avoids rounding to the next kmalloc slab.
In addition this enforces cache line alignment for struct dio to avoid
any false sharing.
Signed-off-by: Andi Kleen <[email protected]>
Acked-by: Jeff Moyer <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
Fix most problems reported by pahole.
There is still a weird 104 byte hole after map_bh. I'm not sure what
causes this.
Signed-off-by: Andi Kleen <[email protected]>
Acked-by: Jeff Moyer <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
This large, but largely mechanic, patch moves all fields in struct dio
that are only used in the submission path into a separate on stack
data structure. This has the advantage that the memory is very likely
cache hot, which is not guaranteed for memory fresh out of kmalloc.
This also gives gcc more optimization potential because it can easier
determine that there are no external aliases for these variables.
The sdio initialization is a initialization now instead of memset.
This allows gcc to break sdio into individual fields and optimize
away unnecessary zeroing (after all the functions are inlined)
Signed-off-by: Andi Kleen <[email protected]>
Acked-by: Jeff Moyer <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>
With the tracing code in there they are far too big to inline.
.text savings compared to a non force inline kernel:
i915_restore_display 4393 12036 +7643
i915_save_display 4295 11459 +7164
i915_handle_error 2979 6666 +3687
i915_driver_irq_handler 2923 5086 +2163
i915_ringbuffer_info 458 1661 +1203
i915_save_vga - 1200 +1200
i915_driver_irq_uninstall 453 1624 +1171
i915_driver_irq_postinstall 913 2078 +1165
ironlake_enable_drps 719 1872 +1153
i915_restore_vga - 1142 +1142
intel_display_capture_error_state 784 2030 +1246
intel_init_emon 719 2016 +1297
and more ...
[AK: these are older numbers, with the new SNB forcewake checks
it will be even worse]
Signed-off-by: Andi Kleen <[email protected]>
Reviewed-by: Daniel Vetter <[email protected]>
Acked-by: Ben Widawsky <[email protected]>
Signed-off-by: Keith Packard <[email protected]>
This shrinks the sizes of a lot of functions in the radeon driver
dramatically.
With a non force inline + -Os kernel this is default anyways.
Cc: David Airlie <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
With this patch I'm only about 50k larger with DRM debugging
enables (why is that enabled by default?!?), and slightly
smaller without.
[airlied: moved r100.c additions to radeon_ring.c]
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
With the dropped inlines gccs starts warning about genuinely unused
functions. Remove r600_bpe_from_format, evergreen_cs_track_validate_cb,
evergreen-cs_packet_next_is_pkt3_nop which are all unused.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
Fixes
evergreen_cs_parse 4080 23124 +19044
and others compared to a non force inline kernel.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
I got a request to make it easier to determine the microcode
update level on Intel CPUs. This patch adds a new "microcode"
field to /proc/cpuinfo.
The microcode level is also outputed on fatal machine checks
together with the other CPUID model information.
I removed the respective code from the microcode update driver,
it just reads the field from cpu_data. Also when the microcode
is updated it fills in the new values too.
I had to add a memory barrier to native_cpuid to prevent it
being optimized away when the result is not used.
This turns out to clean up further code which already got this
information manually. This is done in followon patches.
Signed-off-by: Andi Kleen <[email protected]>
Acked-by: H. Peter Anvin <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Add -M option to report/annotate to pass directly to objdump. This
allows to use -M intel for intel style disassembler syntax, which is
useful for people who are very used to the Intel syntax.
Link: http://lkml.kernel.org/r/[email protected]
[committer note: Add missing Documentation bits, fixup conflicts with 3e6a2a7]
Cc: Frederic Weisbecker <[email protected]>
Cc: Stephane Eranian <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
When a program crashes under perf there is no message about it, unlike
when running it from bash. This can be confusing and lead to wrong
actions during debugging.
Print fatal signals in perf stat/record.
Thanks to Furat Afram for finding the problem originally
Link: http://lkml.kernel.org/r/[email protected]
Cc: Frederic Weisbecker <[email protected]>
Cc: Stephane Eranian <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
rcutree.c defines rcu_cpu_kthread_cpu as int, not unsigned int,
so the extern has to follow that.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>
The externs here didn't agree with the declarations in qos.c.
Better would be probably to move this into a header, but since it's
common practice to have naked externs with sysctls I left it for now.
Cc: [email protected]
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
I ran into a couple of programs which broke with the new Linux 3.0
version. Some of those were binary only. I tried to use LD_PRELOAD to
work around it, but it was quite difficult and in one case impossible
because of a mix of 32bit and 64bit executables.
For example, all kind of management software from HP doesnt work, unless
we pretend to run a 2.6 kernel.
$ uname -a
Linux svivoipvnx001 3.0.0-08107-g97cd98f #1062 SMP Fri Aug 12 18:11:45 CEST 2011 i686 i686 i386 GNU/Linux
$ hpacucli ctrl all show
Error: No controllers detected.
$ rpm -qf /usr/sbin/hpacucli
hpacucli-8.75-12.0
Another notable case is that Python now reports "linux3" from
sys.platform(); which in turn can break things that were checking
sys.platform() == "linux2":
https://bugzilla.mozilla.org/show_bug.cgi?id=664564
It seems pretty clear to me though it's a bug in the apps that are using
'==' instead of .startswith(), but this allows us to unbreak broken
programs.
This patch adds a UNAME26 personality that makes the kernel report a
2.6.40+x version number instead. The x is the x in 3.x.
I know this is somewhat ugly, but I didn't find a better workaround, and
compatibility to existing programs is important.
Some programs also read /proc/sys/kernel/osrelease. This can be worked
around in user space with mount --bind (and a mount namespace)
To use:
wget ftp://ftp.kernel.org/pub/linux/kernel/people/ak/uname26/uname26.c
gcc -o uname26 uname26.c
./uname26 program
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
const __read_mostly is not legal and causes section type conflicts.
That's because the read.mostly section is not read only.
Simply drop the __read_mostly designation.
Signed-off-by: Andi Kleen <[email protected]>
[drop __read_mostly instead of const]
Signed-off-by: Dan Williams <[email protected]>
Some recent benchmarking on btrfs showed that a major scaling bottleneck
on large systems on btrfs is currently the xattr lookup on every write.
Why xattr lookup on every write I hear you ask?
write wants to drop suid and security related xattrs that could set o
capabilities for executables. To do that it currently looks up
security.capability on EVERY write (even for non executables) to decide
whether to drop it or not.
In btrfs this causes an additional tree walk, hitting some per file system
locks and quite bad scalability. In a simple read workload on a 8S
system I saw over 90% CPU time in spinlocks related to that.
Chris Mason tells me this is also a problem in ext4, where it hits
the global mbcache lock.
This patch adds a simple per inode to avoid this problem. We only
do the lookup once per file and then if there is no xattr cache
the decision. All xattr changes clear the flag.
I also used the same flag to avoid the suid check, although
that one is pretty cheap.
A file system can also set this flag when it creates the inode,
if it has a cheap way to do so. This is done for some common file systems
in followon patches.
With this patch a major part of the lock contention disappears
for btrfs. Some testing on smaller systems didn't show significant
performance changes, but at least it helps the larger systems
and is generally more efficient.
v2: Rename is_sgid. add file system helper.
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: Serge E. Hallyn <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Al Viro <[email protected]>
The original INT_MAX is too large, reduce it to
- avoid unnecessarily dirtying/bouncing the cache line
- restore mmap read-around faster on changed access pattern
Background: in the mosbench exim benchmark which does multi-threaded page
faults on shared struct file, the ra->mmap_miss updates are found to cause
excessive cache line bouncing on tmpfs. The ra state updates are needless
for tmpfs because it actually disabled readahead totally
(shmem_backing_dev_info.ra_pages == 0).
Tested-by: Tim Chen <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Wu Fengguang <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
240f62c875 replaced the node_lock with rcu_read_lock, but forgot
to remove the actual lock in the data structure. Remove it here.
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
Commit dde79e005a ("page_cgroup: reduce allocation overhead for
page_cgroup array for CONFIG_SPARSEMEM") added a regression that the
memory cgroup data structures all end up in node 0 because the first
attempt at allocating them would not pass in a node hint. Since the
initialization runs on CPU #0 it would all end up node 0. This is a
problem on large memory systems, where node 0 would lose a lot of
memory.
Change the alloc_pages_exact() to alloc_pages_exact_nid(). This will
still fall back to other nodes if not enough memory is available.
[ RED-PEN: right now it would fall back first before trying
vmalloc_node. Probably not the best strategy ... But I left it like
that for now. ]
Signed-off-by: Andi Kleen <[email protected]>
Reported-by: Doug Nelson
Cc: David Rientjes <[email protected]>
Reviewed-by: Michal Hocko <[email protected]>
Cc: Dave Hansen <[email protected]>
Acked-by: Balbir Singh <[email protected]>
Acked-by: Johannes Weiner <[email protected]>
Reviewed-by: KOSAKI Motohiro <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Right now all RCU walks fall back to reference walk when CONFIG_SECURITY
is enabled, even though just the standard capability module is active.
This is because security_inode_exec_permission unconditionally fails
RCU walks.
Move this decision to the low level security module. This requires
passing the RCU flags down the security hook. This way at least
the capability module and a few easy cases in selinux/smack work
with RCU walks with CONFIG_SECURITY=y
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Eric Paris <[email protected]>
Andi Kleen