* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
powerpc: Fix for getting CPU number in power_save_ppc32_restore()
powerpc: Fix build error with 64K pages and !hugetlbfs
powerpc: Work around gcc's -fno-omit-frame-pointer bug
powerpc: Make sure _etext is after all kernel text
powerpc: Only make kernel text pages of linear mapping executable
powerpc: Fix uninitialised variable in VSX alignment code
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
bnx2x: Accessing un-mapped page
ath9k: Fix TX control flag use for no ACK and RTS/CTS
ath9k: Fix TX status reporting
iwlwifi: fix STATUS_EXIT_PENDING is not set on pci_remove
iwlwifi: call apm stop on exit
iwlwifi: fix Tx cmd memory allocation failure handling
iwlwifi: fix rx_chain computation
iwlwifi: fix station mimo power save values
iwlwifi: remove false rxon if rx chain changes
iwlwifi: fix hidden ssid discovery in passive channels
iwlwifi: W/A for the TSF correction in IBSS
netxen: Remove workaround for chipset quirk
pcnet-cs, axnet_cs: add new IDs, remove dup ID with less info
ixgbe: initialize interrupt throttle rate
net/usb/pegasus: avoid hundreds of diagnostics
tipc: Don't use structure names which easily globally conflict.
Fix a bug and a philosophical decision about who handles errors.
security_context_to_sid_core() was leaking a context in the common case.
This was causing problems on fedora systems which recently have started
making extensive use of this function.
In discussion it was decided that if string_to_context_struct() had an
error it was its own responsibility to clean up any mess it created
along the way.
Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
The allocated RX buffer size was 64 bytes bigger than the PCI mapped
size with no good reason. If the packet was actually using the buffer up
to its limit and if the last 64 bytes of the buffer crossed 4KB boundary
then an unmapped PCI page was accessed. The fix is to use only one
parameter for the buffer size - there is no need to differentiate
between the buffer size and the PCI mapping size since the extra 64
bytes can actually be used by the FW to align the Ethernet payload to
64 bytes.
Also updating the driver version and date
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch sets STATUS_EXIT_PENDING on pci_remove. Otherwise
iwl4965_down may fail to uninitialize the driver.
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Signed-off-by: Mohamed Abbas <mohamed.abbas@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch calls apm stop on exit and suspend. Without this patch
hardware consumes power even after driver is removed or suspended.
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Signed-off-by: Mohamed Abbas <mohamed.abbas@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch "iwlwifi: do not use GFP_DMA in iwl_tx_queue_init" removes
GFP_DMA from allocation tx command buffers. GFP_DMA allows allocation
only for memory under 16M which causes allocation problems
suspend/resume flows.
Using kmalloc is temporal solution and some consistent/coherent
allocation schema will be more correct. Since iwlwifi hardware
supports 64bit address this solution should work on x86 (32 and
64bit) for now.
This patch fixes memory freeing problem in the previous patch.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Ian Schram <ischram@telenet.be>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch fixes rx_chain computation. The code that adjusts number of
rx chains to number supported by HW was missing. Miss configuration
causes firmware error. Note: iwlwifi supports HW with up to 3 RX
chains (2x2, 2x3, 1x2, and 3x3 MIMO). This patch also simplifies the
whole RX chain computation.
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Mohamed Abbas <mohamed.abbas@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch fixes the wrong use MIMO power save values. Our TX was
configured with our MIMO power save values instead of peer's MIMO power
save values, this may affect connectivity. The peer STA/AP may not sense
our traffic at all as it doesn't have all RX chains opened.
Signed-off-by: Ron Rindjunsky <ron.rindjunsky@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Rx chain might change during power save transitions but it doesn't
require sending Full-ROXN command to the firmware. Full-RXON requires
reconnection to an AP and thus affects user experience. The patch
avoids the Full-RXON by removing the rx_chain modification check in
iwl_full_rxon_required function.
Signed-off-by: Mohamed Abbas <mohamed.abbas@intel.com
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This enables sending of direct probes on passive channels, as long as
traffic was detected on that channel. This enables connectivity to
hidden/non broadcasting SSIDs APs on passive channels. Note 5000 HW
declares all 5.2 spectrum as passive.
Signed-off-by: Cahill Ben <ben.m.cahill@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Ron Rindjunsky <ron.rindjunsky@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch is a W/A for the TSF sync issue in IBSS merging. HW is not
capable to sync TSF (it's constantly little behind). This creates
constant IBSS merging upon reception of each beacon, adding and removing
station which in turn creates above 50% packet loss and thus dramatically
degrade the throughput. The W/A simply stops the driver from declaring it
has a reliable TSF value and thus eliminates IBSS merging.
Signed-off-by: Assaf Krauss <assaf.krauss@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The TSC calibration function is still very complicated, but this makes
it at least a little bit less so by moving the PIT part out into a
helper function of its own.
Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove chipset-specific quirk workaround; the workaround caused
unrecoverable DMA lockups when the driver was loaded following a
PXE boot.
Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: Michael Brown <mbrown@fensystems.co.uk>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
This commit dropped the setting of the default interrupt throttle rate.
commit 021230d40a
Author: Ayyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Date: Mon Mar 3 15:03:45 2008 -0800
ixgbe: Introduce MSI-X queue vector code
The following patch adds it back. Without this the default value of 0
causes the performance of this card to be awful. Restoring these to the
default values yields much better performance.
This regression has been around since 2.6.25.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Acked-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: stable@kernel.org [2.6.25 and later]
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Make the "pegasus" driver scream less loudly in the face of
problems as it initializes, avoiding hundreds of messages:
- ratelimit some key error messages
- avoid some spurious diagnostics caused by strange codeflow
And fix one instance of goofy indentation.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
The calculation to get TI_CPU based off of SPRG3 was just plain wrong,
meaning that we were getting garbage for the CPU number on 6xx/G3/G4
based SMP boxes in this code.
Just offset off the stack pointer (to get to thread_info) like all the
other references to TI_CPU do.
This was pointed out by Chen Gong <G.Chen@freescale.com>
[paulus@samba.org - use rlwinm r12,r11,... instead of
rlwinm r12,r1,...; tophys()]
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
HAVE_ARCH_UNMAPPED_AREA and HAVE_ARCH_UNMAPPED_AREA_TOPDOWN must
be defined whenever CONFIG_PPC_MM_SLICES is enabled, not just when
CONFIG_HUGETLB_PAGE is. They used to be always defined together but
this is no longer the case since 3a8247cc2c
("powerpc: Only demote individual slices rather than whole process").
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This bug is causing random crashes
(http://bugzilla.kernel.org/show_bug.cgi?id=11414).
-fno-omit-frame-pointer is only needed on powerpc when -pg is also
supplied, and there is a gcc bug that causes incorrect code generation
on 32-bit powerpc when -fno-omit-frame-pointer is used---it uses stack
locations below the stack pointer, which is not allowed by the ABI
because those locations can and sometimes do get corrupted by an
interrupt.
This ensures that CONFIG_FRAME_POINTER is only selected by ftrace.
When CONFIG_FTRACE is enabled we also pass -mno-sched-epilog to work
around the gcc codegen bug.
Patch based on work by:
Andreas Schwab <schwab@suse.de>
Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Tony Breeds <tony@bakeyournoodle.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This makes core_kernel_text() (and therefore kernel_text_address())
return the correct result. Currently all the __devinit routines (at
least) will not be considered to be kernel text.
This is just a quick fix for 2.6.27 - hopefully we will be able to fix
this better in 2.6.28.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Commit bc033b63bb ("powerpc/mm: Fix
attribute confusion with htab_bolt_mapping()") moved the check for
whether we should make pages of the linear mapping executable from
htab_bolt_mapping into its callers, including htab_initialize.
A side-effect of this is that the decision is now made once for
each contiguous section in the LMB array rather than for each page
individually. This can often mean that the whole of the linear
mapping ends up being executable.
This reverts to the previous behaviour, where individual pages are
checked for being part of the kernel text or not, by moving the check
back down into htab_bolt_mapping.
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This fixes an uninitialised variable in the VSX alignment code. It can
cause warnings from GCC (noticed with gcc-4.1.1). Gcc is actually
correct in this instance, and this bug could cause the alignment
interrupt handler to send a SIGSEGV to the process on a legitimate
access.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Andrew Morton reported a build failure on sparc32, because TIPC
uses names like "struct node" and there is a like named data
structure defined in linux/node.h
This just regexp replaces "struct node*" to "struct tipc_node*"
to avoid this and any future similar problems.
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
ipsec: Fix deadlock in xfrm_state management.
ipv: Re-enable IP when MTU > 68
net/xfrm: Use an IS_ERR test rather than a NULL test
ath9: Fix ath_rx_flush_tid() for IRQs disabled kernel warning message.
ath9k: Incorrect key used when group and pairwise ciphers are different.
rt2x00: Compiler warning unmasked by fix of BUILD_BUG_ON
mac80211: Fix debugfs union misuse and pointer corruption
wireless/libertas/if_cs.c: fix memory leaks
orinoco: Multicast to the specified addresses
iwlwifi: fix 64bit platform firmware loading
iwlwifi: fix apm_stop (wrong bit polarity for FLAG_INIT_DONE)
iwlwifi: workaround interrupt handling no some platforms
iwlwifi: do not use GFP_DMA in iwl_tx_queue_init
net/wireless/Kconfig: clarify the description for CONFIG_WIRELESS_EXT_SYSFS
net: Unbreak userspace usage of linux/mroute.h
pkt_sched: Fix locking of qdisc_root with qdisc_root_sleeping_lock()
ipv6: When we droped a packet, we should return NET_RX_DROP instead of 0
Larry Finger reported at http://lkml.org/lkml/2008/9/1/90:
An ancient laptop of mine started throwing errors from b43legacy when
I started using 2.6.27 on it. This has been bisected to commit bfc0f59
"x86: merge tsc calibration".
The unification of the TSC code adopted mostly the 64bit code, which
prefers PMTIMER/HPET over the PIT calibration.
Larrys system has an AMD K6 CPU. Such systems are known to have
PMTIMER incarnations which run at double speed. This results in a
miscalibration of the TSC by factor 0.5. So the resulting calibrated
CPU/TSC speed is half of the real CPU speed, which means that the TSC
based delay loop will run half the time it should run. That might
explain why the b43legacy driver went berserk.
On the other hand we know about systems, where the PIT based
calibration results in random crap due to heavy SMI/SMM
disturbance. On those systems the PMTIMER/HPET based calibration logic
with SMI detection shows better results.
According to Alok also virtualized systems suffer from the PIT
calibration method.
The solution is to use a more wreckage aware aproach than the current
either/or decision.
1) reimplement the retry loop which was dropped from the 32bit code
during the merge. It repeats the calibration and selects the lowest
frequency value as this is probably the closest estimate to the real
frequency
2) Monitor the delta of the TSC values in the delay loop which waits
for the PIT counter to reach zero. If the maximum value is
significantly different from the minimum, then we have a pretty safe
indicator that the loop was disturbed by an SMI.
3) keep the pmtimer/hpet reference as a backup solution for systems
where the SMI disturbance is a permanent point of failure for PIT
based calibration
4) do the loop iteration for both methods, record the lowest value and
decide after all iterations finished.
5) Set a clear preference to PIT based calibration when the result
makes sense.
The implementation does the reference calibration based on
HPET/PMTIMER around the delay, which is necessary for the PIT anyway,
but keeps separate TSC values to ensure the "independency" of the
resulting calibration values.
Tested on various 32bit/64bit machines including Geode 266Mhz, AMD K6
(affected machine with a double speed pmtimer which I grabbed out of
the dump), Pentium class machines and AMD/Intel 64 bit boxen.
Bisected-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ever since commit 4c563f7669
("[XFRM]: Speed up xfrm_policy and xfrm_state walking") it is
illegal to call __xfrm_state_destroy (and thus xfrm_state_put())
with xfrm_state_lock held. If we do, we'll deadlock since we
have the lock already and __xfrm_state_destroy() tries to take
it again.
Fix this by pushing the xfrm_state_put() calls after the lock
is dropped.
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a bug reported by and diagnosed by Aaron Straus.
This is a regression intruduced into 2.6.26 by
commit adc782dae6
Author: Matt Mackall <mpm@selenic.com>
Date: Tue Apr 29 01:03:07 2008 -0700
random: simplify and rename credit_entropy_store
credit_entropy_bits() does:
spin_lock_irqsave(&r->lock, flags);
...
if (r->entropy_count > r->poolinfo->POOLBITS)
r->entropy_count = r->poolinfo->POOLBITS;
so there is a time window in which this BUG_ON():
static size_t account(struct entropy_store *r, size_t nbytes, int min,
int reserved)
{
unsigned long flags;
BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
/* Hold lock while accounting */
spin_lock_irqsave(&r->lock, flags);
can trigger.
We could fix this by moving the assertion inside the lock, but it seems
safer and saner to revert to the old behaviour wherein
entropy_store.entropy_count at no time exceeds
entropy_store.poolinfo->POOLBITS.
Reported-by: Aaron Straus <aaron@merfinllc.com>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: <stable@kernel.org> [2.6.26.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make PM_QOS and CPU_IDLE play nicer when run with the RT-Preempt kernel.
The purpose of the patch is to remove the spin_lock around the read in the
function pm_qos_requirement - since spinlocks can sleep in -rt and this
function is called from idle.
CPU_IDLE polls the target_value's of some of the pm_qos parameters from
the idle loop causing sleeping locking warnings. Changing the
target_value to an atomic avoids this issue.
Remove the spinlock in pm_qos_requirement by making target_value an atomic
type.
Signed-off-by: mark gross <mgross@linux.intel.com>
Signed-off-by: John Kacur <jkacur@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Update rtc-cmos shutdown handling to leave RTC alarms active, resolving
http://bugzilla.kernel.org/show_bug.cgi?id=11411 on several boards. There
are still some systems where the ACPI event handling doesn't cooperate.
(Possibly related to bugid 11312, reporting the spontaneous disabling of
RTC events.)
Bug 11411 reported that changes to work around some ACPI event issues
broke wake-from-S5 handling, as used for DVR applications. (They like to
power off, then wake later to record programs.)
[yakui.zhao@intel.com: add shutdown for PNP devices]
[dbrownell@users.sourceforge.net: update comments]
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Stefan Bauer <stefan.bauer@cs.tu-chemnitz.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I got this patch through Red Hat's bugzilla from the bug submitter and
patch creator. I have just fixed it up so it applies without fuzz to
upstream kernels.
Original patch and description from Shyam kumar Iyer:
The issue [ibft module not displaying targets with short names] is because
of an offset calculatation error in the iscsi_ibft.c code. Due to this
error directory structure for the target in /sys/firmware/ibft does not
get created and so the initiator is unable to connect to the target.
Note that this bug surfaced only with an name that had a short section at
the end. eg: "iqn.1984-05.com.dell:dell". It did not surface when the
iqn's had a longer section at the end. eg:
"iqn.2001-04.com.example:storage.disk2.sys1.xyz"
So, the eot_offset was calculated such that an extra 48 bytes i.e. the
size of the ibft_header which has already been accounted was subtracted
twice.
This was not evident with longer iqn names because they would overshoot
the total ibft length more than 48 bytes and thus would escape the bug.
Signed-off-by: Shyam Kumar Iyer <shyam_iyer@dell.com>
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Cc: Konrad Rzeszutek <konrad@virtualiron.com>
Cc: Peter Jones <pjones@redhat.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
commit 945185a69d ("rtc: rtc_time_to_tm: use
unsigned arithmetic") changed the some types in rtc_time_to_tm() to
unsigned:
void rtc_time_to_tm(unsigned long time, struct rtc_time *tm)
{
- register int days, month, year;
+ unsigned int days, month, year;
This doesn't work for all cases, because days is checked for < 0 later
on:
if (days < 0) {
year -= 1;
days += 365 + LEAP_YEAR(year);
}
I think the correct fix would be to keep days signed and do an appropriate
cast later on.
Signed-off-by: Jan Altenberg <jan.altenberg@linutronix.de>
Cc: Maciej W. Rozycki <macro@linux-mips.org>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Cc: Dmitri Vorobiev <dmitri.vorobiev@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If there are more then one graphics card handled by the tdfxfb driver the
name of the frame buffer overruns reserved size.
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix memory detection on Voodoo3 cards with SDRAM memory.
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It turns out that event 0x4 merely indcates that a hotkey has been
pressed, not which one. A further query is required in order to determine
the actual keypress. The following patch adds support for that along with
the known keycodes.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
hp-wmi currently changes the RFKill state by altering the struct members
rather than using the dedicated interface, meaning that update events
won't be pushed to userspace. This patch fixes that, along with fixing
the declared type of the WWAN kill switch. It also ensures that rfkill
interfaces are only registered for hardware that exists.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Acked-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Cc: Ivo van Doorn <ivdoorn@gmail.com>
Cc: Dave Young <hidave.darkstar@gmail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Update Documentation/filesystems/proc.txt: it describes the file
auto_msgmni intoduced to enable/disable msgmni automatic recomputing upon
memory add/remove (see thread http://lkml.org/lkml/2008/7/4/27). Also
added a description for msgmni (this filex is only listed in
Documentation/sysctl/kernel.txt).
Signed-off-by: Nadia Derbey <Nadia.Derbey@bull.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Quicklists store pages for each CPU as caches. (Each CPU can cache
node_free_pages/16 pages)
It is used for page table cache. exit() will increase the cache size,
while fork() consumes it.
So for example if an apache-style application runs (one parent and many
child model), one CPU process will fork() while another CPU will process
the middleware work and exit().
At that time, the CPU on which the parent runs doesn't have page table
cache at all. Others (on which children runs) have maximum caches.
QList_max = (#ofCPUs - 1) x Free / 16
=> QList_max / (Free + QList_max) = (#ofCPUs - 1) / (16 + #ofCPUs - 1)
So, How much quicklist memory is used in the maximum case?
This is proposional to # of CPUs because the limit of per cpu quicklist
cache doesn't see the number of cpus.
Above calculation mean
Number of CPUs per node 2 4 8 16
============================== ====================
QList_max / (Free + QList_max) 5.8% 16% 30% 48%
Wow! Quicklist can spend about 50% memory at worst case.
My demonstration program is here
--------------------------------------------------------------------------------
#define _GNU_SOURCE
#include <stdio.h>
#include <errno.h>
#include <stdlib.h>
#include <string.h>
#include <sched.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/wait.h>
#define BUFFSIZE 512
int max_cpu(void) /* get max number of logical cpus from /proc/cpuinfo */
{
FILE *fd;
char *ret, buffer[BUFFSIZE];
int cpu = 1;
fd = fopen("/proc/cpuinfo", "r");
if (fd == NULL) {
perror("fopen(/proc/cpuinfo)");
exit(EXIT_FAILURE);
}
while (1) {
ret = fgets(buffer, BUFFSIZE, fd);
if (ret == NULL)
break;
if (!strncmp(buffer, "processor", 9))
cpu = atoi(strchr(buffer, ':') + 2);
}
fclose(fd);
return cpu;
}
void cpu_bind(int cpu) /* bind current process to one cpu */
{
cpu_set_t mask;
int ret;
CPU_ZERO(&mask);
CPU_SET(cpu, &mask);
ret = sched_setaffinity(0, sizeof(mask), &mask);
if (ret == -1) {
perror("sched_setaffinity()");
exit(EXIT_FAILURE);
}
sched_yield(); /* not necessary */
}
#define MMAP_SIZE (10 * 1024 * 1024) /* 10 MB */
#define FORK_INTERVAL 1 /* 1 second */
main(int argc, char *argv[])
{
int cpu_max, nextcpu;
long pagesize;
pid_t pid;
/* set max number of logical cpu */
if (argc > 1)
cpu_max = atoi(argv[1]) - 1;
else
cpu_max = max_cpu();
/* get the page size */
pagesize = sysconf(_SC_PAGESIZE);
if (pagesize == -1) {
perror("sysconf(_SC_PAGESIZE)");
exit(EXIT_FAILURE);
}
/* prepare parent process */
cpu_bind(0);
nextcpu = cpu_max;
loop:
/* select destination cpu for child process by round-robin rule */
if (++nextcpu > cpu_max)
nextcpu = 1;
pid = fork();
if (pid == 0) { /* child action */
char *p;
int i;
/* consume page tables */
p = mmap(0, MMAP_SIZE, PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
i = MMAP_SIZE / pagesize;
while (i-- > 0) {
*p = 1;
p += pagesize;
}
/* move to other cpu */
cpu_bind(nextcpu);
/*
printf("a child moved to cpu%d after mmap().\n", nextcpu);
fflush(stdout);
*/
/* back page tables to pgtable_quicklist */
exit(0);
} else if (pid > 0) { /* parent action */
sleep(FORK_INTERVAL);
waitpid(pid, NULL, WNOHANG);
}
goto loop;
}
----------------------------------------
When above program which does task migration runs, my 8GB box spends
800MB of memory for quicklist. This is not memory leak but doesn't seem
good.
% cat /proc/meminfo
MemTotal: 7701568 kB
MemFree: 4724672 kB
(snip)
Quicklists: 844800 kB
because
- My machine spec is
number of numa node: 2
number of cpus: 8 (4CPU x2 node)
total mem: 8GB (4GB x2 node)
free mem: about 5GB
- Then, 4.7GB x 16% ~= 880MB.
So, Quicklist can use 800MB.
So, if following spec machine run that program
CPUs: 64 (8cpu x 8node)
Mem: 1TB (128GB x8node)
Then, quicklist can waste 300GB (= 1TB x 30%). It is too large.
So, I don't like cache policies which is proportional to # of cpus.
My patch changes the number of caches
from:
per-cpu-cache-amount = memory_on_node / 16
to
per-cpu-cache-amount = memory_on_node / 16 / number_of_cpus_on_node.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Keiichiro Tokunaga <tokunaga.keiich@jp.fujitsu.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Tested-by: David Miller <davem@davemloft.net>
Acked-by: Mike Travis <travis@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
During the use of a dev_cgroup, we should guarantee the corresponding
cgroup won't be deleted (i.e. via rmdir). This can be done through
css_get(&dev_cgroup->css), but here we can just get and use the dev_cgroup
under rcu_read_lock.
And also remove checking NULL dev_cgroup, it won't be NULL since a task
always belongs to a cgroup.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1. Check if virtual resolution fits into memory.
Otherwise, Linux hangs during panning.
2. When selected use all available memory to
maximize yres_virtual to speed up panning
(previously also xres_virtual was increased).
3. Simplify memory restriction calculations.
Signed-off-by: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We don't change pid_ns->child_reaper when the main thread of the
subnamespace init exits. As Robert Rex <robert.rex@exasol.com> pointed
out this is wrong.
Yes, the re-parenting itself works correctly, but if the reparented task
exits it needs ->parent->nsproxy->pid_ns in do_notify_parent(), and if the
main thread is zombie its ->nsproxy was already cleared by
exit_task_namespaces().
Introduce the new function, find_new_reaper(), which finds the new
->parent for the re-parenting and changes ->child_reaper if needed. Kill
the now unneeded exit_child_reaper().
Also move the changing of ->child_reaper from zap_pid_ns_processes() to
find_new_reaper(), this consolidates the games with ->child_reaper and
makes it stable under tasklist_lock.
Addresses http://bugzilla.kernel.org/show_bug.cgi?id=11391
Reported-by: Robert Rex <robert.rex@exasol.com>
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
zap_pid_ns_processes() sets pid_ns->child_reaper = NULL, this is wrong.
Yes, we have already killed all tasks in this namespace, and sys_wait4()
doesn't see any child. But this doesn't mean ->children list is empty, we
may have EXIT_DEAD tasks which are not visible to do_wait(). In that case
the subsequent forget_original_parent() will crash the kernel because it
will try to re-parent these tasks to the NULL reaper.
Even if there are no childs, it is not good that forget_original_parent()
uses reaper == NULL.
Change the code to set ->child_reaper = init_pid_ns.child_reaper instead.
We could use pid_ns->parent->child_reaper as well, I think this does not
really matter. These EXIT_DEAD tasks are not visible to the new ->parent
after re-parenting, they will silently do release_task() eventually.
Note that we must change ->child_reaper, otherwise
forget_original_parent() will use reaper == father, and in that case we
will hit the (correct) BUG_ON(!list_empty(&father->children)).
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
At91_mci is abusing dma_free_coherent(), which may not be called with IRQs
disabled. I saw "mkfs.ext3" on an MMC card objecting voluminously as each
write completed:
WARNING: at arch/arm/mm/consistent.c:368 dma_free_coherent+0x2c/0x224()
[<c002726c>] (dump_stack+0x0/0x14) from [<c00387d4>] (warn_on_slowpath+0x4c/0x68)
[<c0038788>] (warn_on_slowpath+0x0/0x68) from [<c0028768>] (dma_free_coherent+0x2c/0x224)
r6:00008008 r5:ffc06000 r4:00000000
[<c002873c>] (dma_free_coherent+0x0/0x224) from [<c01918ac>] (at91_mci_irq+0x374/0x420)
[<c0191538>] (at91_mci_irq+0x0/0x420) from [<c0065d9c>] (handle_IRQ_event+0x2c/0x6c)
...
This bug has been around for a LONG time. The MM warning is from late
2005, but the driver merged a year later ... so I'm puzzled why nobody
noticed this before now.
The fix involves noting that this buffer shouldn't be DMA-coherent; it's
just used for normal DMA writes. So replace it with standard kmalloc()
buffering and DMA mapping calls.
This is the quickie fix. A better one would not rely on allocating large
bounce buffers. (Note that dma_alloc_coherent could have failed too, but
that case was ignored... kmalloc is a bit more likely to fail though.)
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Pierre Ossman <drzeus-mmc@drzeus.cx>
Cc: Andrew Victor <linux@maxim.org.za>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Linus Torvalds