On an nForce4-equipped machine with two SATA disk in raid1 setup using dmraid,
we experienced frequent deadlock of the system under high i/o load. 'cat
/dev/zero > ~/zero' was the most reliable way to reproduce them: Randomly
after a few GB, 'cp' would be left in 'D' state along with kjournald and
kmirrord. The functions cp and kjournald were blocked in did vary, but
kmirrord's wchan always pointed to 'mempool_alloc()'. We've seen this pattern
on 2.6.15 and 2.6.17 kernels. http://lkml.org/lkml/2005/4/20/142 indicates
that this problem has been around even before.
So much for the facts, here's my interpretation: mempool_alloc() first tries
to atomically allocate the requested memory, or falls back to hand out
preallocated chunks from the mempool. If both fail, it puts the calling
process (kmirrord in this case) on a private waitqueue until somebody refills
the pool. Where the only 'somebody' is kmirrord itself, so we have a
deadlock.
I worked around this problem by falling back to a (blocking) kmalloc when
before kmirrord would have ended up on the waitqueue. This defeats part of
the benefits of using the mempool, but at least keeps the system running. And
it could be done with a two-line change. Note that mempool_alloc() clears the
GFP_NOIO flag internally, and only uses it to decide whether to wait or return
an error if immediate allocation fails, so the attached patch doesn't change
behaviour in the non-deadlocking case. Path is against current git
(2.6.18-rc4), but should apply to earlier versions as well. I've tested on
2.6.15, where this patch makes the difference between random lockup and a
stable system.
Signed-off-by: Daniel Kobras <kobras@linux.de>
Acked-by: Alasdair G Kergon <agk@redhat.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
In the cleanups of drivers/rtc/s3c-rtc.c, the base address for the
registers got broken. This patch fixes that by ensuring the readb/writeb
are all prefixed with the base returned from ioremap()ing the registers.
Also fix check for valid year range, which was the wrong way around.
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This fixes CCID3 to give much closer performance to RFC4342.
CCID3 is meant to alter sending rate based on RTT and loss.
The performance was verified against:
http://wand.net.nz/~perry/max_download.php
For example I tested with netem and had the following parameters:
Delayed Acks 1, MSS 256 bytes, RTT 105 ms, packet loss 5%.
This gives a theoretical speed of 71.9 Kbits/s. I measured across three
runs with this patch set and got 70.1 Kbits/s. Without this patchset the
average was 232 Kbits/s which means Linux can't be used for CCID3 research
properly.
I also tested with netem turned off so box just acting as router with 1.2
msec RTT. The performance with this is the same with or without the patch
at around 30 Mbit/s.
Signed off by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: David S. Miller <davem@davemloft.net>
The bridge-netfilter code will overwrite memory if there is not
headroom in the skb to save the header. This first showed up when
using Xen with sky2 driver that doesn't allocate the extra space.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
[DCCP]: Introduce dccp_rx_hist_find_entry
[DCCP]: Introduces follows48 function
[DCCP]: Update contact details and copyright
[DCCP]: Fix typo
[IPV6]: Segmentation offload not set correctly on TCP children
[CONNECTOR]: Add userspace example code into Documentation/connector/
This adds a new function dccp_rx_hist_find_entry.
Signed off by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds a new function to see if two sequence numbers follow each
other.
Signed off by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes a small typo in net/dccp/libs/packet_history.c
Signed off by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP over IPV6 would incorrectly inherit the GSO settings.
This would cause kernel to send Tcp Segmentation Offload packets for
IPV6 data to devices that can't handle it. It caused the sky2 driver
to lock http://bugzilla.kernel.org/show_bug.cgi?id=7050
and the e1000 would generate bogus packets. I can't blame the
hardware for gagging if the upper layers feed it garbage.
This was a new bug in 2.6.18 introduced with GSO support.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
I was asked several times to include userspace example code into
Documentation, so if there is no policy against it, consider attached patch
for 2.6.18. This program works with included Documentation/connector/cn_test.c
connector module.
Signed-off-by: Evgeniy Polyakov <johnpol@2ka.mipt.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current sun disklabel code uses a signed int for the sector count.
When partitions larger than 1 TB are used, the cast to a sector_t causes
the partition sizes to be invalid:
# cat /proc/paritions | grep sdan
66 112 2146435072 sdan
66 115 9223372036853660736 sdan3
66 120 9223372036853660736 sdan8
This patch switches the sector count to an unsigned int to fix this.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
It moves the smp_procesors_ready variable to sun4d_smp.c only.
Signed-off-by: Krzysztof Helt (krzysztof.h1@wp.pl)
Signed-off-by: David S. Miller <davem@davemloft.net>
smp_setup_cpu_possible_map() needs to run after paging_init()
so that the in-kernel device tree is setup.
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
The existing unusual_devs entry for the UCR-61S2B appears to have too
wide a revision range. It matches at least one device that doesn't
respond to the initialization sequence. Perhaps the sequence needs to
be updated, or perhaps something else can be done. For now, this patch
(as764) restricts the range to include only the revision mentioned in
the original comment.
This resolves (for now!) Bugzilla entry #6950.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch removes support for a clone of Nokia DKU-5 cable made by Ours
Technology Inc, as it turned out that the cable does not use the pl2303
chip, but OTI-6858 chip which is not compatible with the pl2303.
Signed-off-by: Tomasz Kazmierczak <tomek.fizyk@op.pl>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This was pointed out by Adrian Bunk <bunk@stusta.de>, as found by the Coverity Checker.
Cc: Adrian Bunk <bunk@stusta.de>
Cc: Oliver Bock <o.bock@fh-wolfenbuettel.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Removes an unused kerneldoc entry from pci_match_device and
put the others into correct order.
Signed-off-by: Henrik Kretzschmar <henne@nachtwindheim.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Here is a patch against the CPCI hotplug core to fix up PCI resource
assignment such that things will actually work when a hot inserted
device is enabled. I mentioned this patch to you way back in April at
ELC, but am only now out from under things enough to clean it up and
submit it. I've basically cribbed the corresponding code from
shpchp_pci.c, so there are no big surprises. If it's still possible, I
wouldn't mind this going into 2.6.18, but it wouldn't be the end of the
world if it went into 2.6.19.
Signed-off-by: Scott Murray <scottm@somanetworks.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
- add the ICH6(R) LPC to the ICH6 ACPI quirks. currently only the ICH6-M
is handled. [ PCI_DEVICE_ID_INTEL_ICH6_1 is the ICH6-M LPC, ICH6_0 is
the ICH6(R) ]
- remove the wrong quirk calling asus_hides_smbus_lpc() for ICH6. the
register modified in asus_hides_smbus_lpc() has a different meaning in
ICH6.
Signed-off-by: Daniel Ritz <daniel.ritz@gmx.ch>
Acked-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
On i386 PCI mmconfig forgets the bus number when setting the fallback_slots
bits which means fallback to conf1 only works for bus 0.
Signed-off-by: Daniel Ritz <daniel.ritz@gmx.ch>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
there was a change in 2.6.17 which affected the order in which the PCI
access methods are probed. this gives regressions on some machines with
broken BIOS. the problem is that PCBIOS sometimes reports last bus wrong,
leaving cardbus non-funcational. previously those system worked fine with
direct access.
The patch changes the PCI init code to have PCBIOS as last fallback, yet
the PCBIOS code still has to run first to set pcibios_last_bus to the value
reported by the BIOS. this is needed in case legacy PCI probing
(arch/i386/pci/legacy.c) is used to detect peer busses. using direct
access if available fixes the cardbus problems.
Signed-off-by: Daniel Ritz <daniel.ritz@gmx.ch>
Cc: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch contains 2 sets of fixes for the abituguru:
1) Much improved timeout handling, drasticly reducing the amount of
timeout errors on some motherboards
2) Fix the exit paths in the bank1 sensor type detect code to always
restore the original settings even on an error. Without this our
special test settings could remain seriously confusing the system
BIOS's setup menu.
Both are very much related and are must haves, to avoid messing up the
uguru CMOS settings.
Detailed changes:
- Much improved timeout / wait for status handling. Many thanks to Sunil
Kumar, for all his testing, ideas and patches! The code now first busy
waits, polling the uguru for the expected status as this usually
succeeds pretty quickly (within 90 reads). To avoid unnecessary CPU burn
in timeout conditions, the amount of busy waiting has been halved from
previous versions (120 tries instead of 250). This is not a problem,
because this version goes to sleep after 120 attemps for 1 jiffy and
then tries again, it does this sleep and try again 5 times before
finally giving up. This (almost?) completly removes the timeout errors
some people have seen regulary. Apparently some older uguru versions
sometimes are distracted for a (relatively) long time. This solves this.
- These timeout errors not only occur in the sending address part of
reading the uguru but also in the wait for read state, so errors in
this state are now handled as retryable just like send address state
errors and are only logged and reported to userspace if 3 executive
tries fail.
- Fix a very nasty bug in the bank1 sensor type detection code, where it
would not restore the original settings in any of the error paths!
- Since not successfully restoring the original settings can seriously
confuse the system BIOS (hang when entering the relevant setup menu),
we now try restoring them 3 times before giving up.
Signed-off-by: Hans de Goede <j.w.r.degoede@hhs.nl>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The tps65010.c driver in the main tree never got updated with
build fixes since the last batch of I2C driver changes; and the
genirq trigger flags were updated wierdly too.
This also includes a minor tweak to reduce the frequency used to
poll for unplug-the-AC-power on the TPS chips that don't provide
relevant IRQs. It _would_ be nice to sense whether there's even
a battery, but that'd normally be an HDQ/1-wire interface to a
smart battery, and such APIs aren't standardized.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The callers of scsi_send_eh_cmnd are setting the cmnd buffer,
and then scsi_send_eh_cmnd is copying that updated buffer to
the old_cmnd variable. Then after the command runs, we end up
copying that old_cmnd var which has the new cmnd to the scsi
command buffer. When this command gets recent, all types of fun
things happen like getting TUR or START_STOP commands with
data and scatterlists.
This patch made against scsi-rc-fixes, has the callers of
scsi_send_eh_cmnd pass in the command so scsi_send_eh_cmnd
can do the right thing. This should go into 2.6.18 since this
fixes a regression added when we removed some of the scsi_cmnd
fields and replaced them with local variables.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Software must explicitely re-enable extended firmware tracing
after any ISP abort condition.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Original code attempts to retry PLOGIs to fcports that are
FCP_TARGETs only. If the driver never performed a successful
PLOGI/PRLI, the port-type would never be assigned, and the
relogin logic would silently drop the request (and thus the port
would not be recognized and registered).
The fix is relatively straightforward, drop the FCP_TARGET-only
check.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
There's a problem where sg is executing a ->nopage operation on a
compound page, it actually calls get_page() on the first page in the
compound rather than the page which is being mapped. The fix is to
select the correct page by indexing into the compound.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The check in open_exec() for inode->i_mode & 0111 has been made
redundant by the fix to permission().
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 1d3741c5d991686699f100b65b9956f7ee7ae0ae commit)
The check in prepare_binfmt() for inode->i_mode & 0111 is redundant,
since open_exec() will already have done that.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 822dec482ced07af32c378cd936d77345786572b commit)
Currently, the access() call will return incorrect information on NFS if
there exists an ACL that grants execute access to the user on a regular
file. The reason the information is incorrect is that the VFS overrides
this execute access in open_exec() by checking (inode->i_mode & 0111).
This patch propagates the VFS execute bit check back into the generic
permission() call.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 64cbae98848c4c99851cb0a405f0b4982cd76c1e commit)
This is needed in order to handle any NFS4ERR_DELAY errors that might be
returned by the server. It also ensures that we map the NFSv4 errors before
they are returned to userland.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 71c12b3f0abc7501f6ed231a6d17bc9c05a238dc commit)
Check the bounds of length specifiers more thoroughly in the XDR decoding of
NFS4 readdir reply data.
Currently, if the server returns a bitmap or attr length that causes the
current decode point pointer to wrap, this could go undetected (consider a
small "negative" length on a 32-bit machine).
Also add a check into the main XDR decode handler to make sure that the amount
of data is a multiple of four bytes (as specified by RFC-1014). This makes
sure that we can do u32* pointer subtraction in the NFS client without risking
an undefined result (the result is undefined if the pointers are not correctly
aligned with respect to one another).
Signed-Off-By: David Howells <dhowells@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 5861fddd64a7eaf7e8b1a9997455a24e7f688092 commit)
Neil Brown observed that the current limit of 32 bytes isn't enough to hold two
ip addresses and the rest of the stuff we're putting in it, so it's often
truncated to the point where it's unlikely to be unique. This can cause
spurious CLID_INUSE's from the server.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from fc8c17ec251e984ab3df9182ed097aa5b577c915 commit)
Some hardware uses port 664 for its hardware-based IPMI listener. Teach
the RPC client to avoid using that port by raising the default minimum port
number to 665.
Test plan:
Find a mainboard known to use port 664 for IPMI; enable IPMI; mount NFS
servers in a tight loop.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 58e8cb3a035d22fc386e1c53a5d98c3f219530fb commit)
The problem is that we may be caching writes that would extend the file and
create a hole in the region that we are reading. In this case, we need to
detect the eof from the server, ensure that we zero out the pages that
are part of the hole and mark them as up to date.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 856b603b01b99146918c093969b6cb1b1b0f1c01 commit)
nlm_traverse_files() is not allowed to hold the nlm_file_mutex while calling
nlm_inspect file, since it may end up calling nlm_release_file() when
releaseing the blocks.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from e558d3cde986e04f68afe8c790ad68ef4b94587a commit)
rpc_unlink() and rpc_rmdir() will dput the dentry reference for you.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from a05a57effa71a1f67ccbfc52335c10c8b85f3f6a commit)
A prior call to rpc_depopulate() by rpc_rmdir() on the parent directory may
have already called simple_unlink() on this entry.
Add the same check to rpc_rmdir(). Also remove a redundant call to
rpc_close_pipes() in rpc_rmdir.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 0bbfb9d20f6437c4031aa3bf9b4d311a053e58e3 commit)
Make it take a dentry argument instead of a path
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 648d4116eb2509f010f7f34704a650150309b3e7 commit)
I'm trying to speeding up mkdir(2) for network file systems. A typical
mkdir(2) calls two inode_operations: lookup and mkdir. The lookup
operation would fail with ENOENT in common case. I think it is unnecessary
because the subsequent mkdir operation can check it. In case of creat(2),
lookup operation is called with the LOOKUP_CREATE flag, so individual
filesystem can omit real lookup. e.g. nfs_lookup().
Here is a sample patch which uses LOOKUP_CREATE and O_EXCL on mkdir,
symlink and mknod. This uses the gadget for creat(2).
And here is the result of a benchmark on NFSv3.
mkdir(2) 10,000 times:
original 50.5 sec
patched 29.0 sec
Signed-off-by: ASANO Masahiro <masano@tnes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from fab7bf44449b29f9d5572a5dd8adcf7c91d5bf0f commit)
nfs_wb_page() waits on request completion and, as a result, is not safe to be
called from nfs_release_page() invoked by VM scanner as part of GFP_NOFS
allocation. Fix possible deadlock by analyzing gfp mask and refusing to
release page if __GFP_FS is not set.
Signed-off-by: Nikita Danilov <danilov@gmail.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
(cherry picked from 374d969debfb290bafcb41d28918dc6f7e43ce31 commit)
The subsystem check in the PAV code is incorrect, it enables PAV
per device instead of per subsystem.
Signed-off-by: Horst Hummel <horst.hummel@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
vt6420 has super-fragile SCR registers which can hang the whole
machine if accessed with the wrong timings. This patch makes sata_via
use SCR registers only during probing and with the same timings as
before (pre new EH), which is proven to work.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
