[PATCH] md: fix minor error in raid10 read-balancing calculation.
'this_sector' is a virtual (array) address while 'head_position' is a physical
(device) address, so substraction doesn't make any sense. devs[slot].addr
should be used instead of this_sector.
However, this patch doesn't make much practical different to the read
balancing due to the effects of later code.
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Don't just irritate all other kernel developers. Fix the users first,
then you can re-introduce the must-check infrastructure to avoid new
cases creeping in.
Daniel Ritz [Thu, 8 Sep 2005 22:57:14 +0000 (00:57 +0200)]
[PATCH] Update PCI IOMEM allocation start
This fixes the problem with "Averatec 6240 pcmcia_socket0: unable to
apply power", which was due to the CardBus IOMEM register region being
allocated at an address that was actually inside the RAM window that had
been reserved for video frame-buffers in an UMA setup.
The BIOS _should_ have marked that region reserved in the e820 memory
descriptor tables, but did not.
It is fixed by rounding up the default starting address of PCI memory
allocations, so that we leave a bigger gap after the final known memory
location. The amount of rounding depends on how big the unused memory
gap is that we can allocate IOMEM from.
Clean up timer initialization by introducing DEFINE_TIMER a'la
DEFINE_SPINLOCK. Build and boot-tested on x86. A similar patch has been
been in the -RT tree for some time.
[PATCH] FUSE: don't allow restarting of system calls
This patch removes ability to interrupt and restart operations while there
hasn't been any side-effect.
The reason: applications. There are some apps it seems that generate
signals at a fast rate. This means, that if the operation cannot make
enough progress between two signals, it will be restarted for ever. This
bug actually manifested itself with 'krusader' trying to open a file for
writing under sshfs. Thanks to Eduard Czimbalmos for the report.
The problem can be solved just by making open() uninterruptible, because in
this case it was the truncate operation that slowed down the progress. But
it's better to solve this by simply not allowing interrupts at all (except
SIGKILL), because applications don't expect file operations to be
interruptible anyway. As an added bonus the code is simplified somewhat.
Don't change mtime/ctime/atime to local time on read/write. Rather invalidate
file attributes, so next stat() will force a GETATTR call. Bug reported by
Ben Grimm.
Make data caching behavior selectable on a per-open basis instead of
per-mount. Compatibility for the old mount options 'kernel_cache' and
'direct_io' is retained in the userspace library (version 2.4.0-pre1 or
later).
[PATCH] fuse: transfer readdir data through device
This patch removes a long lasting "hack" in FUSE, which used a separate
channel (a file descriptor refering to a disk-file) to transfer directory
contents from userspace to the kernel.
The patch adds three new operations (OPENDIR, READDIR, RELEASEDIR), which
have semantics and implementation exactly maching the respective file
operations (OPEN, READ, RELEASE).
This simplifies the directory reading code. Also disk space is not
necessary, which can be important in embedded systems.
This patch adds support for the "direct_io" mount option of FUSE.
When this mount option is specified, the page cache is bypassed for
read and write operations. This is useful for example, if the
filesystem doesn't know the size of files before reading them, or when
any kind of caching is harmful.
[PATCH] FUSE: tighten check for processes allowed access
This patch tightens the check for allowing processes to access non-privileged
mounts. The rational is that the filesystem implementation can control the
behavior or get otherwise unavailable information of the filesystem user. If
the filesystem user process has the same uid, gid, and is not suid or sgid
application, then access is safe. Otherwise access is not allowed unless the
"allow_other" mount option is given (for which policy is controlled by the
userspace mount utility).
Thanks to everyone linux-fsdevel, especially Martin Mares who helped uncover
problems with the previous approach.
With the help of the readpages() operation multiple reads are bundled
together and sent as a single request to userspace. This can improve
reading performace.
This patch adds miscellaneous mount options to the FUSE filesystem.
The following mount options are added:
o default_permissions: check permissions with generic_permission()
o allow_other: allow other users to access files
o allow_root: allow root to access files
o kernel_cache: don't invalidate page cache on open
The most important difference between orinary filesystems and FUSE is
the fact, that the filesystem data/metadata is provided by a userspace
process run with the privileges of the mount "owner" instead of the
kernel, or some remote entity usually running with elevated
privileges.
The security implication of this is that a non-privileged user must
not be able to use this capability to compromise the system. Obvious
requirements arising from this are:
- mount owner should not be able to get elevated privileges with the
help of the mounted filesystem
- mount owner should not be able to induce undesired behavior in
other users' or the super user's processes
- mount owner should not get illegitimate access to information from
other users' and the super user's processes
These are currently ensured with the following constraints:
1) mount is only allowed to directory or file which the mount owner
can modify without limitation (write access + no sticky bit for
directories)
2) nosuid,nodev mount options are forced
3) any process running with fsuid different from the owner is denied
all access to the filesystem
1) and 2) are ensured by the "fusermount" mount utility which is a
setuid root application doing the actual mount operation.
3) is ensured by a check in the permission() method in kernel
I started thinking about doing 3) in a different way because Christoph
H. made a big deal out of it, saying that FUSE is unacceptable into
mainline in this form.
The suggested use of private namespaces would be OK, but in their
current form have many limitations that make their use impractical (as
discussed in this thread).
Suggested improvements that would address these limitations:
- implement shared subtrees
- allow a process to join an existing namespace (make namespaces
first-class objects)
- implement the namespace creation/joining in a PAM module
With all that in place the check of owner against current->fsuid may
be removed from the FUSE kernel module, without compromising the
security requirements.
Suid programs still interesting questions, since they get access even
to the private namespace causing some information leak (exact
order/timing of filesystem operations performed), giving some
ptrace-like capabilities to unprivileged users. BTW this problem is
not strictly limited to the namespace approach, since suid programs
setting fsuid and accessing users' files will succeed with the current
approach too.
This patch brings the now out-of-date Documentation/filesystems/vfs.txt
back to life. Thanks to Carsten Otte, Trond Myklebust, and Anton
Altaparmakov for their help on updating this documentation.
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Pekka Enberg [Fri, 9 Sep 2005 20:10:16 +0000 (13:10 -0700)]
[PATCH] update kfree, vfree, and vunmap kerneldoc
This patch clarifies NULL handling of kfree() and vfree(). I addition,
wording of calling context restriction for vfree() and vunmap() are changed
from "may not" to "must not."
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Update the hacking guide, before CONFIG_PREEMPT_RT goes in and it needs
rewriting again.
Changes include modernization of quotes, removal of most references to
bottom halves (some mention required because we still use bh in places to
mean softirq).
It would be nice to have a discussion of sparse and various annotations.
Please send patches straight to akpm.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (authored) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This set of two patches add support for the framebuffer of the Samsung S3C2410
ARM SoC. This driver was started about one year ago and is now used on iPAQ
h1930/h1940, Acer n30 and probably other s3c2410-based machines I'm not aware
of. I've also heard yesterday that it's working also on iPAQ rx3715/rx3115
(s3c2440-based machines).
Signed-Off-By: Arnaud Patard <arnaud.patard@rtp-net.org> Signed-off-by: Antonino Daplas <adaplas@pol.net> Signed-off-by: Ben Dooks <ben@trinity.fluff.org> Cc: Russell King <rmk@arm.linux.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add ddc/i2c support for i810fb. This will allow the driver to get display
information, especially for monitors with fickle timings. The i2c support
depends on CONFIG_FB_I810_GTF.
Changed __init* to __devinit*
Signed-off-by: Antonino Daplas <adaplas@pol.net> Signed-off-by: Alexander Nyberg <alexn@telia.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
[PATCH] fbcon: Break up bit_putcs into its component functions
The function bit_putcs() in drivers/video/console/bitblit.c is becoming large.
Break it up into its component functions (bit_putcs_unaligned and
bit_putcs_aligned).
Incorporated fb_pad_aligned_buffer() optimization by Roman Zippel.
Richard Purdie [Fri, 9 Sep 2005 20:10:03 +0000 (13:10 -0700)]
[PATCH] pxafb: Add hsync time reporting hook
To solve touchscreen interference problems devices like the Sharp Zaurus
SL-C3000 need to know the length of the horitzontal sync pulses. This patch
adds a hook to pxafb so the touchscreen driver can function correctly.
Signed-Off-By: Richard Purdie <rpurdie@rpsys.net> Signed-off-by: Antonino Daplas <adaplas@pol.net> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
[PATCH] nvidiafb: Fixed mirrored characters in big endian machines
nvidiafb_imageblit converts the bitdata stream from big_endian to little
endian. This produces mirrored characters when machine is big_endian. Do not
endian convert on big endian machines.
[PATCH] nvidiafb: Use CVT to get mode for digital displays
If no EDID block is probed, if the display is digital and if no mode option is
specified by the user, get the timings by CVT instead of using the global mode
database.
[PATCH] fbdev: Add VESA Coordinated Video Timings (CVT) support
The Coordinated Video Timings (CVT) is the latest standard approved by VESA
concerning video timings generation. It addresses the limitation of GTF which
is designed mainly for CRT displays. CRT's have a high blanking requirement
(as much as 25% of the horizontal frame length) which artificially increases
the pixelclock. Digital displays, on the other hand, needs to conserve the
pixelclock as much as possible. The GTF also does not take into account the
different aspect ratios in its calculation.
The new function added is fb_find_mode_cvt(). It is called by fb_find_mode()
if it recognizes a mode option string formatted for CVT. The format is:
The 'M' tells the function to calculate using CVT. On it's own, it will
compute a timing for CRT displays at 60Hz. If the 'R' is specified, 'reduced
blanking' computation will be used, best for flatpanels. The 'i' and the 'm'
is for 'interlaced mode' and 'with margins' respectively.
To determine if CVT was used, check for dmesg for something like this:
where: pix - product of xres and yres, in MB
M - is a CVT mode
n - the aspect ratio (3 - 4:3; 4 - 5:4; 9 - 16:9, 15:9; A - 16:10)
-R - reduced blanking
Paul Mackerras [Mon, 5 Sep 2005 23:31:03 +0000 (09:31 +1000)]
[PATCH] PCI: Small rearrangement of PCI probing code
This patch makes some small rearrangements of the PCI probing code in
order to make it possible for arch code to set up the PCI tree
without needing to duplicate code from the PCI layer unnecessarily.
PPC64 will use this to set up the PCI tree from the Open Firmware
device tree, which we need to do on logically-partitioned pSeries
systems.
Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Rajesh Shah [Wed, 17 Aug 2005 00:32:04 +0000 (17:32 -0700)]
[PATCH] PCI: Fix PCI bus mastering enable problem in pciehp
Martin Franc reported that the pciehp driver was not enabling bus
master capability on his hot-plugged card. pciehprm_enable_card()
was updating the PCI command register only if _HPP indicated a
value for SERR or PERR that was different from the current setting.
I don't have hardware that reproduces this problem, but Martin
reports that this patch fixes the problem for him.
Signed-off-by: Rajesh Shah <rajesh.shah@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
From rzarev@its.caltech.edu Tue Sep 6 18:29:50 2005
Date: Tue, 6 Sep 2005 13:39:32 -0700 (PDT)
From: Rumen Ivanov Zarev <rzarev@its.caltech.edu>
Message-Id: <200509062039.j86KdWMr014934@inky.its.caltech.edu>
To: gregkh@suse.de
Subject: PCI: Unhide SMBus on Compaq Evo N620c Cc: linux-kernel@vger.kernel.org
Trivial patch against 2.6.13 to unhide SMBus on Compaq Evo N620c laptop using
Intel 82855PM chipset.
Daniel Burcaw [Fri, 9 Sep 2005 20:04:59 +0000 (13:04 -0700)]
[PATCH] radeonfb: Only request resources we need
This patch changes radeon to request only resources 0 and 2 instead of all
3. This works around problems with some setups where BAR 1 (IO BAR) has
not been assigned by the firmware since it's not used on the machine and
the kernel fails to assign something to it due to the card being between a
P2P bridge that was configured without an IO range at all.
This typically fixes radeonfb on some Apple Xserve G5 machines
Signed-off-by: Daniel Burcaw <dburcaw@terrasoftsolutions.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
[PATCH] framebuffer: bit_putcs() optimization for 8x* fonts
This trivial patch gives a performance boost to the framebuffer console
Constructing the bitmaps that are given to the bitblit functions of the
framebuffer drivers is time consuming. Here we avoide a call to the slow
fb_pad_aligned_buffer(). The patch replaces that call with a simple but
much more efficient bytewise copy.
The kernel spends a significant time at this place if you use 8x* fonts.
Every pixel displayed on your screen is prepared here.
Some benchmark results:
Displaying a file of 2000 lines with 160 characters each takes 889 ms
system time using cyblafb on my system (I´m using a 1280x1024 video mode,
resulting in a 160x64 character console)
Displaying the same file with the enclosed patch applied to 2.6.13 only
takes 760 ms system time, saving 129 ms or 14.5%.
Font widths other than 8 are not affected.
The advantage and correctness of this patch should be obvious.
[PATCH] framebuffer: new driver for cyberblade/i1 graphics core
This is a framebuffer driver for the Cyberblade/i1 graphics core.
Currently tridenfb claims to support the cyberblade/i1 graphics core. This
is of very limited truth. Even vesafb is faster and provides more working
modes and a much better quality of the video signal. There is a great
number of bugs in tridentfb ... but most often it is impossible to decide
if these bugs are real bugs or if fixing them for the cyberblade/i1 core
would break support for one of the other supported chips.
Tridentfb seems to be unmaintained,and documentation for most of the
supported chips is not available. So "fixing" cyberblade/i1 support inside
of tridentfb was not an option, it would have caused numerous
if(CYBERBLADEi1) else ... cases and would have rendered the code to be
almost unmaintainable.
A first version of this driver was published on 2005-07-31. A fix for a
bug reported by Jochen Hein was integrated as well as some changes
requested by Antonino A. Daplas.
A message has been added to tridentfb to inform current users of tridentfb
to switch to cyblafb if the cyberblade/i1 graphics core is detected.
This patch is one logical change, but because of the included documentation
it is bigger than 70kb. Therefore it is not sent to lkml and
linux-fbdev-devel,
Ian Romanick [Fri, 9 Sep 2005 20:04:42 +0000 (13:04 -0700)]
[PATCH] matroxfb: read MGA PInS data on PowerPC
This updates the matroxfb code so that it can find the PInS data embedded
in the BIOS on PowerPC cards. The process for finding the data is
different on OpenFirmware cards than on x86 cards, and the code for doing
so was missing.
After patching, building, installing, and booting a kernel, you should grep
for "PInS" in /var/log/messages. You should see two messages in the log:
PInS data found at offset XXXXX
PInS memtype = X
On the GXT135p card I get "31168" and "5". The first value is irrelevant,
but it's presence lets me know that the PInS data was actually found. On a
GXT130p, the second value should be 3. Since I don't have access to that
hardware, if someone can verify that, I will submit a follow-on patch that
rips out all the memtype parameter stuff.
Signed-off-by: Ian Romanick <idr@us.ibm.com> Signed-off-by: Petr Vandrovec <vandrove@vc.cvut.cz> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
On a vc resize, the contents of the old screen buffer are transferred to the
new screenbuffer. If the new screenbuffer is smaller than the old one, only
the contents from the bottom are copied to new. If the contents of the old
buffer are located at the top, then the contents will not be copied to the new
buffer resulting in a blank screen.
This bug will happen only if the vc in question is not in the foreground.
Doing an fbset -a or con2fbmap will trigger this bug.
To fix this problem, base the start of the copy from the location of the
current cursor. If the cursor is near the top of the buffer, copy the
contents at the top, and if the cursor is near the bottom of the buffer, then
copy the contents at the bottom. In the unlikely case where the new row size
is greater than 2x smaller than the old one, and the cursor is in the middle,
copy 1/2 screenful from the top and bottom of the cursor position.
[PATCH] fbcon: Saner 16-color to 4-color conversion
Currently, the default linux 16-colors are converted to 4-colors by simply
dividing the values by 4. However, this is not necessarily correct since the
first 4 colors are converted to black, rendering them invisible.
So, for black, no conversion; for light colors, convert to gray, for normal
text color, no conversion, and for bright colors, convert to intense white.
[PATCH] fbdev: Fix greater than 1 bit monochrome color handling
Currently, fbcon assumes that the visual FB_VISUAL_MONO* is always 1 bit.
According to Geert, there are old hardware where it's possible to have
monochrome at 8-bit, but has only 2 colors, black - 0x00 and white - 0xff.
Fix color handlers (fb_get_color_depth, and get_color) for this special case.
[PATCH] fbdev: Resurrect hooks to get EDID from firmware
For the i386, code is already present in video.S that gets the EDID from the
video BIOS. Make this visible so drivers can also use this data as fallback
when i2c does not work.
To ensure that the EDID block is returned for the primary graphics adapter
only, by check if the IORESOURCE_ROM_SHADOW flag is set.
David Vrabel [Fri, 9 Sep 2005 20:04:32 +0000 (13:04 -0700)]
[PATCH] fbdev: geode updates
Geode framebuffer driver updates:
- Local mode list (taken from modedb.c) containing only relevant modes.
This also makes the driver work as a module.
- Make it a PCI driver (from James Simmons <jsimmons@infradead.org>).
- A few other minor cosmetic bits and pieces.
Signed-off-by: David Vrabel <dvrabel@arcom.com> Signed-off-by: James Simmons <jsimmons@infradead.org> Cc: "Antonino A. Daplas" <adaplas@pol.net> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
James Simmons [Fri, 9 Sep 2005 20:04:31 +0000 (13:04 -0700)]
[PATCH] fbdev: prevent drivers that have hardware cursors from calling software cursor code
This patch removes drivers that have hardware cursors from calling the
software cursor code. Also if the driver sets a no hardware cursor flag
then the driver reports a error it someone attempts to use the cursor.
Signed-off-by: James Simmons <jsimmons@infradead.org> Cc: "Antonino A. Daplas" <adaplas@pol.net> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add rudimentary support by manipulating the VGA registers. However, not
all vesa modes are VGA compatible, so VGA compatiblity is checked first.
Only 2 levels are supported, powerup and powerdown.
[PATCH] v9fs: fix handling of malformed 9P messages
This patch attempts to do a better job of cleaning up after detecting errors
on the transport. This should also improve error reporting on broken
connections to servers.
Signed-off-by: Latchesar Ionkov <lucho@ionkov.net> Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
LANL reported some issues with random crashes during mount of legacy protocol
servers (9P2000 versus 9P2000.u) -- crash was always happening in readlink
(which should never happen in legacy mode). Added some sanity conditionals to
the get_inode code which should prevent the errors LANL was seeing. Code
tested benign through regression.
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
[PATCH] v9fs: Clean-up vfs_inode and setattr functions
Cleanup code in v9fs vfs_inode as suggested by Alexey Dobriyan. Did some
major revamping of the v9fs setattr code to remove unnecessary allocations and
clean up some dead-code.
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
V9FS is a distributed file system for Linux which provides an
implementation of the Plan 9 resource sharing protocol 9P. It can be
used to share all sorts of resources: static files, synthetic file servers
(such as /proc or /sys), devices, and application file servers (such as
FUSE).
BACKGROUND
Plan 9 (http://plan9.bell-labs.com/plan9) is a research operating
system and associated applications suite developed by the Computing
Science Research Center of AT&T Bell Laboratories (now a part of
Lucent Technologies), the same group that developed UNIX , C, and C++.
Plan 9 was initially released in 1993 to universities, and then made
generally available in 1995. Its core operating systems code laid the
foundation for the Inferno Operating System released as a product by
Lucent Bell-Labs in 1997. The Inferno venture was the only commercial
embodiment of Plan 9 and is currently maintained as a product by Vita
Nuova (http://www.vitanuova.com). After updated releases in 2000 and
2002, Plan 9 was open-sourced under the OSI approved Lucent Public
License in 2003.
The Plan 9 project was started by Ken Thompson and Rob Pike in 1985.
Their intent was to explore potential solutions to some of the
shortcomings of UNIX in the face of the widespread use of high-speed
networks to connect machines. In UNIX, networking was an afterthought
and UNIX clusters became little more than a network of stand-alone
systems. Plan 9 was designed from first principles as a seamless
distributed system with integrated secure network resource sharing.
Applications and services were architected in such a way as to allow
for implicit distribution across a cluster of systems. Configuring an
environment to use remote application components or services in place
of their local equivalent could be achieved with a few simple command
line instructions. For the most part, application implementations
operated independent of the location of their actual resources.
Commercial operating systems haven't changed much in the 20 years
since Plan 9 was conceived. Network and distributed systems support is
provided by a patchwork of middle-ware, with an endless number of
packages supplying pieces of the puzzle. Matters are complicated by
the use of different complicated protocols for individual services,
and separate implementations for kernel and application resources.
The V9FS project (http://v9fs.sourceforge.net) is an attempt to bring
Plan 9's unified approach to resource sharing to Linux and other
operating systems via support for the 9P2000 resource sharing
protocol.
V9FS HISTORY
V9FS was originally developed by Ron Minnich and Maya Gokhale at Los
Alamos National Labs (LANL) in 1997. In November of 2001, Greg Watson
setup a SourceForge project as a public repository for the code which
supported the Linux 2.4 kernel.
About a year ago, I picked up the initial attempt Ron Minnich had
made to provide 2.6 support and got the code integrated into a 2.6.5
kernel. I then went through a line-for-line re-write attempting to
clean-up the code while more closely following the Linux Kernel style
guidelines. I co-authored a paper with Ron Minnich on the V9FS Linux
support including performance comparisons to NFSv3 using Bonnie and
PostMark - this paper appeared at the USENIX/FREENIX 2005
conference in April 2005:
( http://www.usenix.org/events/usenix05/tech/freenix/hensbergen.html ).
CALL FOR PARTICIPATION/REQUEST FOR COMMENTS
Our 2.6 kernel support is stabilizing and we'd like to begin pursuing
its integration into the official kernel tree. We would appreciate any
review, comments, critiques, and additions from this community and are
actively seeking people to join our project and help us produce
something that would be acceptable and useful to the Linux community.
STATUS
The code is reasonably stable, although there are no doubt corner cases
our regression tests haven't discovered yet. It is in regular use by several
of the developers and has been tested on x86 and PowerPC
(32-bit and 64-bit) in both small and large (LANL cluster) deployments.
Our current regression tests include fsx, bonnie, and postmark.
It was our intention to keep things as simple as possible for this
release -- trying to focus on correctness within the core of the
protocol support versus a rich set of features. For example: a more
complete security model and cache layer are in the road map, but
excluded from this release. Additionally, we have removed support for
mmap operations at Al Viro's request.
PERFORMANCE
Detailed performance numbers and analysis are included in the FREENIX
paper, but we show comparable performance to NFSv3 for large file
operations based on the Bonnie benchmark, and superior performance for
many small file operations based on the PostMark benchmark. Somewhat
preliminary graphs (from the FREENIX paper) are available
(http://v9fs.sourceforge.net/perf/index.html).
RESOURCES
The source code is available in a few different forms:
The user-level server is available from either the Plan 9 distribution
or from http://v9fs.sf.net
Other support applications are still being developed, but preliminary
version can be downloaded from sourceforge.
Documentation on the protocol has historically been the Plan 9 Man
pages (http://plan9.bell-labs.com/sys/man/5/INDEX.html), but there is
an effort under way to write a more complete Internet-Draft style
specification (http://v9fs.sf.net/rfc).
There are a couple of mailing lists supporting v9fs, but the most used
is v9fs-developer@lists.sourceforge.net -- please direct/cc your
comments there so the other v9fs contibutors can participate in the
conversation. There is also an IRC channel: irc://freenode.net/#v9fs
This part of the patch contains Documentation, Makefiles, and configuration
file changes.
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With the use of RCU in files structure, the look-up of files using fds can now
be lock-free. The lookup is protected by rcu_read_lock()/rcu_read_unlock().
This patch changes the readers to use lock-free lookup.
Signed-off-by: Maneesh Soni <maneesh@in.ibm.com> Signed-off-by: Ravikiran Thirumalai <kiran_th@gmail.com> Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Patch to eliminate struct files_struct.file_lock spinlock on the reader side
and use rcu refcounting rcuref_xxx api for the f_count refcounter. The
updates to the fdtable are done by allocating a new fdtable structure and
setting files->fdt to point to the new structure. The fdtable structure is
protected by RCU thereby allowing lock-free lookup. For fd arrays/sets that
are vmalloced, we use keventd to free them since RCU callbacks can't sleep. A
global list of fdtable to be freed is not scalable, so we use a per-cpu list.
If keventd is already handling the current cpu's work, we use a timer to defer
queueing of that work.
Since the last publication, this patch has been re-written to avoid using
explicit memory barriers and use rcu_assign_pointer(), rcu_dereference()
premitives instead. This required that the fd information is kept in a
separate structure (fdtable) and updated atomically.
In order for the RCU to work, the file table array, sets and their sizes must
be updated atomically. Instead of ensuring this through too many memory
barriers, we put the arrays and their sizes in a separate structure. This
patch takes the first step of putting the file table elements in a separate
structure fdtable that is embedded withing files_struct. It also changes all
the users to refer to the file table using files_fdtable() macro. Subsequent
applciation of RCU becomes easier after this.
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com> Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The performance improvement is very significant.
We are getting killed by the cacheline bouncing of the files_struct
lock here. Writes on ramdisk (ext2) seems to vary just too
much to get any meaningful number.
Also, With Tridge's thread_perf test on a 4(8)-way (HT) P4 xeon system :
2.6.12-rc5-vanilla :
Running test 'readwrite' with 8 tasks
Threads 0.34 +/- 0.01 seconds
Processes 0.16 +/- 0.00 seconds
2.6.12-rc5-fd :
Running test 'readwrite' with 8 tasks
Threads 0.17 +/- 0.02 seconds
Processes 0.17 +/- 0.02 seconds
I repeated the measurements on ramfs (as opposed to ext2 on ramdisk in
the earlier measurement) and I got more consistent results from tiobench :
2.6.12-rc5-vanilla
--------------------
Running test 'readwrite' with 4 tasks
Threads 0.20 +/- 0.02 seconds
Processes 0.16 +/- 0.01 seconds
2.6.12-rc5-fd
--------------------
Running test 'readwrite' with 4 tasks
Threads 0.18 +/- 0.04 seconds
Processes 0.16 +/- 0.01 seconds
The benefits are huge (upto ~60%) in some cases on x86 primarily
due to the atomic operations during acquisition of ->file_lock
and cache line bouncing in fast path. ppc64 benefits are modest
due to LL/SC based locking, but still statistically significant.
This patch:
RCU head initilizer no longer needs the head varible name since we don't use
list.h lists anymore.
projects / linux-2.6 / log