11 years agorandom32: seeding improvement
Stephen Hemminger [Wed, 30 Jul 2008 23:29:19 +0000]
random32: seeding improvement

The rationale is:
   * use u32 consistently
   * no need to do LCG on values from (better) get_random_bytes
   * use more data from get_random_bytes for secondary seeding
   * don't reduce state space on srandom32()
   * enforce state variable initialization restrictions

Note: the second paper has a version of random32() with even longer period
and a version of random64() if needed.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

11 years agobridge: send correct MTU value in PMTU (revised)
Simon Wunderlich [Wed, 30 Jul 2008 23:27:55 +0000]
bridge: send correct MTU value in PMTU (revised)

When bridging interfaces with different MTUs, the bridge correctly chooses
the minimum of the MTUs of the physical devices as the bridges MTU.  But
when a frame is passed which fits through the incoming, but not through
the outgoing interface, a "Fragmentation Needed" packet is generated.

However, the propagated MTU is hardcoded to 1500, which is wrong in this
situation.  The sender will repeat the packet again with the same frame
size, and the same problem will occur again.

Instead of sending 1500, the (correct) MTU value of the bridge is now sent
via PMTU.  To achieve this, the corresponding rtable structure is stored
in its net_bridge structure.

Modified to get rid of fake_net_device as well.

Signed-off-by: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agoMerge branch 'upstream-davem' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
David S. Miller [Wed, 30 Jul 2008 22:44:30 +0000]
Merge branch 'upstream-davem' of /linux/kernel/git/jgarzik/netdev-2.6

11 years agodrivers/net/ehea/ehea_main.c: Release mutex in error handling code
Julia Lawall [Mon, 21 Jul 2008 07:57:26 +0000]
drivers/net/ehea/ehea_main.c: Release mutex in error handling code

The mutex is released on a successful return, so it would seem that it
should be released on an error return as well.

The semantic patch finds this problem is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression l;
@@

mutex_lock(l);
... when != mutex_unlock(l)
    when any
    when strict
(
if (...) { ... when != mutex_unlock(l)
+   mutex_unlock(l);
    return ...;
}
|
mutex_unlock(l);
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agosh_eth: Add support of SH7763 to sh_eth
Nobuhiro Iwamatsu [Mon, 30 Jun 2008 02:08:17 +0000]
sh_eth: Add support of SH7763 to sh_eth

SH7763 has Ethernet core same as SH7710/SH7712.
Positions of some registry are different, but the basic part is the same.
I add support of ethernet of sh7763 to sh_eth.

Signed-off-by: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agofix NE2000 linkage error
Mikael Pettersson [Wed, 30 Jul 2008 11:44:55 +0000]
fix NE2000 linkage error

Trying to build with CONFIG_NE2000=m fails with:

  scripts/mod/modpost   -o /tmp/tmp/linux-2.6.27-rc1/Module.symvers    -S     -s
ERROR: "NS8390_init" [drivers/net/ne.ko] undefined!

This is because the split of 8390 into pausing and non-pausing
versions was incompletely propagated to ne.c. This fixes it.

Signed-off-by: Mikael Pettersson <mikpe@it.uu.se>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agonet: Make "networking" one-click deselectable.
Robert P. J. Day [Wed, 30 Jul 2008 10:14:01 +0000]
net: Make "networking" one-click deselectable.

Use a menuconfig directive to make all of networking support one-click
deselectable from the top-level menu.

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agoipv6: Fix useless proc net sockstat6 removal
Daniel Lezcano [Wed, 30 Jul 2008 10:12:31 +0000]
ipv6: Fix useless proc net sockstat6 removal

This call is no longer needed, sockstat6 is per namespace so it is
removed at the namespace subsystem destruction.

Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agotcp: MD5: Use MIB counter instead of warning for MD5 mismatch.
David S. Miller [Wed, 30 Jul 2008 10:03:15 +0000]
tcp: MD5: Use MIB counter instead of warning for MD5 mismatch.

From a report by Matti Aarnio, and preliminary patch by Adam Langley.

Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agopkt_sched: Fix OOPS on ingress qdisc add.
David S. Miller [Wed, 30 Jul 2008 09:37:46 +0000]
pkt_sched: Fix OOPS on ingress qdisc add.

Bug report from Steven Jan Springl:

Issuing the following command causes a kernel oops:
tc qdisc add dev eth0 handle ffff: ingress

The problem mostly stems from all of the special case handling of
ingress qdiscs.

So, to fix this, do the grafting operation the same way we do for TX
qdiscs.  Which means that dev_activate() and dev_deactivate() now do
the "qdisc_sleeping <--> qdisc" transitions on dev->rx_queue too.

Future simplifications are possible now, mainly because it is
impossible for dev_queue->{qdisc,qdisc_sleeping} to be NULL.  There
are NULL checks all over to handle the ingress qdisc special case
that used to exist before this commit.

Signed-off-by: David S. Miller <davem@davemloft.net>

11 years ago[netdrvr] wd: fix build breakage with new NS8390p API
Alan Cox [Tue, 8 Jul 2008 18:35:13 +0000]
[netdrvr] wd: fix build breakage with new NS8390p API

From: Alan Cox <alan@redhat.com>

Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agodm9601: don't do usb transfers of data on stack
Peter Korsgaard [Mon, 14 Jul 2008 07:07:32 +0000]
dm9601: don't do usb transfers of data on stack

dm_{read,write}() were doing USB transfers of data on stack, which isn't
allowed. Fix it by kmalloc'ing a temporary buffer.
Clean up the error handling for short transfers while we're at it.

Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoniu: Fix error checking in niu_ethflow_to_class.
Andreas Schwab [Wed, 30 Jul 2008 06:59:20 +0000]
niu: Fix error checking in niu_ethflow_to_class.

The callers of niu_ethflow_to_class expect zero as error, but it returns
-1 instead.

Signed-off-by: Andreas Schwab <schwab@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agoIPv6: datagram_send_ctl() should exit immediately when an error occured
Miao Xie [Wed, 30 Jul 2008 06:57:58 +0000]
IPv6: datagram_send_ctl() should exit immediately when an error occured

When an error occured, datagram_send_ctl() should exit immediately rather than
continue to run the for loop. Otherwise, the variable err might be changed and
the error might be hidden.

Fix this bug by using "goto" instead of "break".

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
David S. Miller [Wed, 30 Jul 2008 04:51:00 +0000]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

11 years agoS2io: fix statistics flush after a MTU change
Breno Leitao [Tue, 22 Jul 2008 19:27:20 +0000]
S2io: fix statistics flush after a MTU change

On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics.  This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.

To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.

Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoenc28j60: don't specify (wrong) IRQ type
David Brownell [Fri, 25 Jul 2008 00:47:56 +0000]
enc28j60: don't specify (wrong) IRQ type

Recent changes to the IRQ framework have made passing the wrong
trigger type to request_irq() become a fatal error.  In the case
of the enc28j60 driver, it stopped working in my test harness.

(Specifically:  the signal detects "pin change" events, both edges,
not just falling edges.  Similarly, other boards might route it
through an inverter.  Trigger type are board-specific.)

This fixes that problem by the usual fix of expecting board setup
code to have set up the correct IRQ trigger type.  The best known
example of that being x86 setup.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoblackfin_mac: unneeded assignment
Stephen Hemminger [Fri, 25 Jul 2008 18:46:46 +0000]
blackfin_mac: unneeded assignment

skb->dev is set by eth_type_trans already.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoMerge branch 'for-2.6.27' of git://git.marvell.com/mv643xx_eth into upstream-fixes
Jeff Garzik [Tue, 29 Jul 2008 22:23:13 +0000]
Merge branch 'for-2.6.27' of git://git.marvell.com/mv643xx_eth into upstream-fixes

11 years agocxgb3: Allow 64KB firmware images.
Steve Wise [Sat, 26 Jul 2008 20:40:56 +0000]
cxgb3: Allow 64KB firmware images.

Starting with FW version 7.0, the driver needs to allow larger images.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoBlackfin EMAC Driver: Functional power management support
Michael Hennerich [Sun, 27 Jul 2008 14:45:05 +0000]
Blackfin EMAC Driver: Functional power management support

Reprogram MAC address after resume from Suspend Mem
(Blackfin Hibernate looses all CORE and SYSTEM register content)

Signed-off-by: Michael Hennerich <michael.hennerich@analog.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoBlackfin EMAC Driver: enable TXDWA new feature for new silicon (rev > 0.2)
Bryan Wu [Sun, 27 Jul 2008 14:45:04 +0000]
Blackfin EMAC Driver: enable TXDWA new feature for new silicon (rev > 0.2)

Signed-off-by: Bryan Wu <cooloney@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoBlackfin EMAC Driver: add proper __devinit/__devexit markings
Mike Frysinger [Sun, 27 Jul 2008 14:45:03 +0000]
Blackfin EMAC Driver: add proper __devinit/__devexit markings

Signed-off-by: Mike Frysinger <vapier.adi@gmail.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agoforcedeth bug fix: realtek phy 8211c errata
Ayaz Abdulla [Fri, 25 Jul 2008 19:31:29 +0000]
forcedeth bug fix: realtek phy 8211c errata

This patch adds support for the realtek 8211c phy. The driver must
perform a hardware reset of the phy due to an errata where the phy could
not detect the link.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>

11 years agomac80211: fix mesh beaconing
Luis Carlos Cobo [Tue, 29 Jul 2008 17:59:31 +0000]
mac80211: fix mesh beaconing

This patch fixes mesh beaconing, which was broken by "mac80211: revamp
beacon configuration".

Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoPS3: gelic: use unsigned long for irqflags
Vegard Nossum [Tue, 29 Jul 2008 11:31:47 +0000]
PS3: gelic: use unsigned long for irqflags

The semantic patch I used was this:

@@
expression lock;
identifier flags;
expression subclass;
@@

- unsigned int flags;
+ unsigned long flags;

...

<+...

(
 spin_lock_irqsave(lock, flags)
|
 _spin_lock_irqsave(lock)
|
 spin_unlock_irqrestore(lock, flags)
|
 _spin_unlock_irqrestore(lock, flags)
|
 read_lock_irqsave(lock, flags)
|
 _read_lock_irqsave(lock)
|
 read_unlock_irqrestore(lock, flags)
|
 _read_unlock_irqrestore(lock, flags)
|
 write_lock_irqsave(lock, flags)
|
 _write_lock_irqsave(lock)
|
 write_unlock_irqrestore(lock, flags)
|
 _write_unlock_irqrestore(lock, flags)
|
 spin_lock_irqsave_nested(lock, flags, subclass)
|
 _spin_lock_irqsave_nested(lock, subclass)
|
 spin_unlock_irqrestore(lock, flags)
|
 _spin_unlock_irqrestore(lock, flags)
|
 _raw_spin_lock_flags(lock, flags)
|
 __raw_spin_lock_flags(lock, flags)
)

...+>

This patch was generated using the Coccinelle framework.

Cc: Masakazu Mokuno <mokuno@sm.sony.co.jp>
Cc: Julia Lawall <julia@diku.dk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: fix cfg80211 hooks for master interface
Johannes Berg [Tue, 29 Jul 2008 11:22:52 +0000]
mac80211: fix cfg80211 hooks for master interface

The master interface is a virtual interface that is registered
to mac80211, changing that does not seem like a good idea at
the moment. However, since it has no sdata, we cannot accept
any configuration for it. This patch makes the cfg80211 hooks
reject any such attempt.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agonl80211: fix dump callbacks
Johannes Berg [Tue, 29 Jul 2008 11:22:51 +0000]
nl80211: fix dump callbacks

Julius Volz pointed out that the dump callbacks in nl80211 were
broken and fixed one of them. This patch fixes the other three
and also addresses the TODOs there.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Julius Volz <juliusv@google.com>
Cc: Thomas Graf <tgraf@suug.ch>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: partially fix skb->cb use
Johannes Berg [Tue, 29 Jul 2008 09:32:07 +0000]
mac80211: partially fix skb->cb use

This patch fixes mac80211 to not use the skb->cb over the queue step
from virtual interfaces to the master. The patch also, for now,
disables aggregation because that would still require requeuing,
will fix that in a separate patch. There are two other places (software
requeue and powersaving stations) where requeue can happen, but that is
not currently used by any drivers/not possible to use respectively.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agortl8187: Improve wireless statistics for RTL8187B
Larry Finger [Tue, 29 Jul 2008 03:25:08 +0000]
rtl8187: Improve wireless statistics for RTL8187B

Wireless statistics produced by the RTL8187B driver are not particularly
informative about the strength of the received signal. From the data sheet
provided by Realtek, I discovered that certain parts of the RX header
should have the information necessary to calculate signal quality and
strength. With testing, it became clear that most of these quantities were
very jittery - only the AGC correlated with the signals expected from nearby
AP's. As a result, the quality and strength are derived from the agc value.
The scaling has been determined so that the numbers are close to those
obtained by b43 under the same conditions. The results are qualitatively
correct.

Statistics derived for the RTL8187 have not been changed.

The RX header variables have been renamed to match the quantites described
in the Realtek data sheet.

Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agortl8187: Fix for TX sequence number problem
Larry Finger [Tue, 29 Jul 2008 03:08:18 +0000]
rtl8187: Fix for TX sequence number problem

"mac80211: fix TX sequence numbers" broke rtl8187. This
patch makes the same kind of fix that was done for rt2x00. Note that
this code will have to be reworked for proper sequence numbers on beacons.
In addition, the sequence number has been placed in the hardware state,
not the vif state.

Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: append CONFIG_ to MAC80211_VERBOSE_PS_DEBUG in net/mac80211/tx.c.
Rami Rosen [Thu, 24 Jul 2008 07:40:37 +0000]
mac80211: append CONFIG_ to MAC80211_VERBOSE_PS_DEBUG in net/mac80211/tx.c.

In net/mac80211/tx.c, there are some #ifdef which checks
MAC80211_VERBOSE_PS_DEBUG
(which in fact is never set) instead of
CONFIG_MAC80211_VERBOSE_PS_DEBUG, as should be.

This patch replaces MAC80211_VERBOSE_PS_DEBUG with
CONFIG_MAC80211_VERBOSE_PS_DEBUG in these #ifdef commands in
net/mac80211/tx.c.

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: fix sparse integer as NULL pointer warning
Harvey Harrison [Thu, 24 Jul 2008 01:36:38 +0000]
mac80211: fix sparse integer as NULL pointer warning

drivers/net/wireless/mac80211_hwsim.c:503:20: warning: Using plain integer as NULL pointer

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agodrivers/net/wireless/iwlwifi/iwl-led.c: printk fix
Andrew Morton [Wed, 23 Jul 2008 06:50:04 +0000]
drivers/net/wireless/iwlwifi/iwl-led.c: printk fix

ia64:

drivers/net/wireless/iwlwifi/iwl-led.c: In function `iwl_get_blink_rate':
drivers/net/wireless/iwlwifi/iwl-led.c:271: warning: long long int format, s64 arg (arg 6)
drivers/net/wireless/iwlwifi/iwl-led.c:271: warning: long long int format, u64 arg (arg 7)

We do not know what type the architecture uses to impement u64 and s64,
hence we must cast the variables for printing.

Cc: Tomas Winkler <tomas.winkler@intel.com>
Cc: John W. Linville <linville@tuxdriver.com>
Cc: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: return correct error return from ieee80211_wep_init
Jeremy Fitzhardinge [Mon, 14 Jul 2008 19:52:08 +0000]
mac80211: return correct error return from ieee80211_wep_init

Return the proper error code rather than a hard-coded ENOMEM from
ieee80211_wep_init.  Also, print the error code on failure.

Signed-off-by: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: tx, use dev_kfree_skb_any for beacon_get
Jiri Slaby [Mon, 14 Jul 2008 10:43:23 +0000]
mac80211: tx, use dev_kfree_skb_any for beacon_get

Use dev_kfree_skb_any(); instead of dev_kfree_skb();, since
ieee80211_beacon_get function might be called from atomic.
(It's in a fail path.)

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Johannes Berg <johannes@sipsolutions.net>
Cc: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Clear queue entry flags during initialization
Ivo van Doorn [Mon, 21 Jul 2008 17:06:02 +0000]
rt2x00: Clear queue entry flags during initialization

When the queues are being initialized the entry flags fields must be
reset to 0. When this does not happen some entries might still be
marked as "occupied" after an ifdown & ifup cycle which would trigger
errors when the entry is being accessed:

phy0 -> rt2x00queue_write_tx_frame: Error - Arrived at non-free entry in the non-full queue 0.
Please file bug report to http://rt2x00.serialmonkey.com.

This also fixes the mac80211 warning:

------------[ cut here ]------------
WARNING: at net/mac80211/tx.c:1238 ieee80211_master_start_xmit+0x30a/0x350 [mac80211]()

which was triggered by the queue error.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Force full register config after start()
Ivo van Doorn [Mon, 21 Jul 2008 09:52:44 +0000]
rt2x00: Force full register config after start()

rt2x00 will only perform configuration changes from
mac80211 when the configuration option has changed.
This means it keeps track of the current active configuration
and will check these values when the config() callback function
is used.

However this causes breakage when the interface has been
brought down and up again, since all stored active values
aren't reset while the registers might have.
This is for example the case with rt61pci antenna registers which
will jump to invalid values when the interface has been started.

To make sure a full configuration takes place after the start()
callback function, a new flag is added which will be checked
during config() and skips the "what's changed" phase.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agob43legacy: Release mutex in error handling code
Julia Lawall [Mon, 21 Jul 2008 09:29:34 +0000]
b43legacy: Release mutex in error handling code

The mutex is released on a successful return, so it would seem that it
should be released on an error return as well.

The semantic patch finds this problem is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression l;
@@

mutex_lock(l);
... when != mutex_unlock(l)
    when any
    when strict
(
if (...) { ... when != mutex_unlock(l)
+   mutex_unlock(l);
    return ...;
}
|
mutex_unlock(l);
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Michael Buesch <mb@bu3sch.de>
Cc: stable <stable@kernel.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agodrivers/net/wireless/ipw2100.c: Release mutex in error handling code
Julia Lawall [Mon, 21 Jul 2008 07:58:11 +0000]
drivers/net/wireless/ipw2100.c: Release mutex in error handling code

The mutex is released on a successful return, so it would seem that it
should be released on an error return as well.

The semantic patch finds this problem is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression l;
@@

mutex_lock(l);
... when != mutex_unlock(l)
    when any
    when strict
(
if (...) { ... when != mutex_unlock(l)
+   mutex_unlock(l);
    return ...;
}
|
mutex_unlock(l);
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Fix memleak when RTS/CTS fails
Ivo van Doorn [Sun, 20 Jul 2008 16:03:58 +0000]
rt2x00: Fix memleak when RTS/CTS fails

When sending the RTS/CTS frame fails, we should
free the skb buffer which was created.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Fix QOS sequence counting
Ivo van Doorn [Sun, 20 Jul 2008 16:03:38 +0000]
rt2x00: Fix QOS sequence counting

When IEEE80211_TX_CTL_ASSIGN_SEQ is not set,
the driver should disable hardware sequence counting
to make sure the mac80211 provided counter is used.
This fixes QOS sequence counting, since that is one
of the cases where mac80211 provides a seperate
sequence counter.

By moving the sequence counting code to rt2x00queue
we make sure that _all_ frames get the sequence counter,
including RTS/CTS and Beacon frames.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Fix the beacon length bug
Iwo Mergler [Sat, 19 Jul 2008 14:17:40 +0000]
rt2x00: Fix the beacon length bug

When setting up a beacon template, the length of the beacon is
calculated with the assumption that the SKB already contains
the Tx descriptor. In the case of beacons it doesn't.

This patch undoes the damage by adding the Tx descriptor length
to the beacon length. This is safe, because the shortest possible
beacon is longer than the Tx header.

Signed-off-by: Iwo Mergler <Iwo@call-direct.com.au>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Large vendor requests for rt73usb firmware upload and beacons
Iwo Mergler [Sat, 19 Jul 2008 14:17:16 +0000]
rt2x00: Large vendor requests for rt73usb firmware upload and beacons

Switches rt73usb to use large vendor requests for firmware
and beacons. This also fixes the garbled beacon bug.

Signed-off-by: Iwo Mergler <Iwo@call-direct.com.au>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Support for large vendor requests
Iwo Mergler [Sat, 19 Jul 2008 14:16:54 +0000]
rt2x00: Support for large vendor requests

Adds an extra rt2x00 vendor request function to support register
transfers beyond the CSR_CACHE_SIZE / USB packet size limit. This
is useful for firmware uploads, beacon templates and keys, all
of which are to large to do with a single USB request.

Signed-off-by: Iwo Mergler <Iwo@call-direct.com.au>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Fix EIFS timing value
Ivo van Doorn [Sat, 19 Jul 2008 14:16:12 +0000]
rt2x00: Fix EIFS timing value

Olivier reported a difference between the EIFS
values used in the legacy driver and the one in
the rt2x00 drivers.

In rt2x00 the value was
( SIFS + (8 * (IEEE80211_HEADER + ACK_SIZE)) )
which comes down to 314us while the legacy driver uses the value 364us

This was caused because EIFS is: SIFS + DIFS + AckTime
This patch will fix this by adding the DIFS by the above value,
and creating a SHORT_EIFS define which uses the SHORT_DIFS.

Reported-by: Olivier Cornu <o.cornu@gmail.com>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agort2x00: Remove duplicate declaration
Ivo van Doorn [Sat, 19 Jul 2008 14:15:42 +0000]
rt2x00: Remove duplicate declaration

rt2x00queue_free_skb() was declared twice.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoath5k: don't enable MSI, we cannot handle it yet
Pavel Roskin [Fri, 18 Jul 2008 16:56:59 +0000]
ath5k: don't enable MSI, we cannot handle it yet

MSI is a nice thing, but we cannot enable it without changing the
interrupt handler.  If we do it, we break MSI capable hardware,
specifically AR5006 chipset.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Cc: stable <stable@kernel.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoath5k: fix recursive locking in ath5k_beacon_update
Bob Copeland [Fri, 18 Jul 2008 15:11:21 +0000]
ath5k: fix recursive locking in ath5k_beacon_update

ath5k_beacon_update takes sc->lock upon entry.  However, it is only
called from within ath5k_config_interface, which already holds the lock.
Remove the unnecessary locking from ath5k_beacon_update.

Signed-off-by: Bob Copeland <me@bobcopeland.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoiwlwifi: small compile warnings without CONFIG_IWLWIFI_DEBUG
Denis V. Lunev [Fri, 18 Jul 2008 06:56:12 +0000]
iwlwifi: small compile warnings without CONFIG_IWLWIFI_DEBUG

CC [M] drivers/net/wireless/iwlwifi/iwl-scan.o
drivers/net/wireless/iwlwifi/iwl-scan.c: In function 'iwl_rx_scan_complete_notif':
drivers/net/wireless/iwlwifi/iwl-scan.c:274: warning: unused variable 'scan_notif'

Signed-off-by: Denis V. Lunev <den@openvz.org>
Acked-by: Tomas Winkler <tomasw@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoiwlwifi: compilation error when CONFIG_IWLWIFI_DEBUG is not set
Denis V. Lunev [Thu, 17 Jul 2008 09:19:24 +0000]
iwlwifi: compilation error when CONFIG_IWLWIFI_DEBUG is not set

CC [M] drivers/net/wireless/iwlwifi/iwl-rfkill.o
drivers/net/wireless/iwlwifi/iwl-led.c: In function 'iwl_led_brightness_set':
drivers/net/wireless/iwlwifi/iwl-led.c:198: error: 'led_type_str' undeclared (first use in this function)
drivers/net/wireless/iwlwifi/iwl-led.c:198: error: (Each undeclared identifier is reported only once
drivers/net/wireless/iwlwifi/iwl-led.c:198: error: for each function it appears in.)

The problem is that led_type_str is defined under CONFIG_IWLWIFI_DEBUG
while IWL_DEBUG is a static inline function in this case. Replace it
with macro.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agolibertas: check bounds and only use decimal for sysfs persistent features.
Brian Cavagnolo [Wed, 16 Jul 2008 19:15:26 +0000]
libertas: check bounds and only use decimal for sysfs persistent features.

Some persistent settings were using hex and others decimal.  In some cases,
values were set in hex but reported in decimal.  Confusing.

Signed-off-by: Brian Cavagnolo <brian@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoiwl-3945: add #ifdef CONFIG_IWL3945_LEDS to avoid compile warning.
Rami Rosen [Wed, 16 Jul 2008 13:39:56 +0000]
iwl-3945: add #ifdef CONFIG_IWL3945_LEDS to avoid compile warning.

When building the wireless-next-2.6 tree with CONFIG_IWL3945 (for building
iwl-3945 driver) and where CONFIG_IWL3945_LEDS is not set,
we get this warning:
drivers/net/wireless/iwlwifi/iwl-3945.c: In function
'iwl3945_pass_packet_to_mac80211':
drivers/net/wireless/iwlwifi/iwl-3945.c:633: warning: unused variable 'hdr'

This patch adds #ifdef to iwl3945_pass_packet_to_mac80211() to avoid this
warning. (The variable 'hdr' is used only if CONFIG_IWL3945_LEDS is set)

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agorfkill: yet more minor kernel-doc fixes
Henrique de Moraes Holschuh [Tue, 22 Jul 2008 00:18:22 +0000]
rfkill: yet more minor kernel-doc fixes

For some stupid reason, I sent and old version of the patch minor kernel
doc-fix patch, and it got merged before I noticed the problem. This is an
incremental fix on top.

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agorfkill: mutex fixes
Henrique de Moraes Holschuh [Tue, 22 Jul 2008 00:18:20 +0000]
rfkill: mutex fixes

There are two mutexes in rfkill:

rfkill->mutex, which protects some of the fields of a rfkill struct, and is
also used for callback serialization.

rfkill_mutex, which protects the global state, the list of registered
rfkill structs and rfkill->claim.

Make sure to use the correct mutex, and to not miss locking rfkill->mutex
even when we already took rfkill_mutex.

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agorfkill: document the rfkill struct locking (v2)
Henrique de Moraes Holschuh [Tue, 22 Jul 2008 00:18:19 +0000]
rfkill: document the rfkill struct locking (v2)

Reorder fields in struct rfkill and add comments to make it clear
which fields are protected by rfkill->mutex.

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agorfkill: fix led-trigger unregister order in error unwind
Henrique de Moraes Holschuh [Tue, 22 Jul 2008 00:18:18 +0000]
rfkill: fix led-trigger unregister order in error unwind

rfkill needs to unregister the led trigger AFTER a call to
rfkill_remove_switch(), otherwise it will not update the LED state,
possibly leaving it ON when it should be OFF.

To make led-trigger unregistering safer, guard against unregistering a
trigger twice, and also against issuing trigger events to a led trigger
that was unregistered.  This makes the error unwind paths more resilient.

Refer to "rfkill: Register LED triggers before registering switch".

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Cc: Michael Buesch <mb@bu3sch.de>
Cc: Dmitry Baryshkov <dbaryshkov@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agorfkill: document rfkill_force_state as required (v2)
Henrique de Moraes Holschuh [Tue, 22 Jul 2008 00:18:17 +0000]
rfkill: document rfkill_force_state as required (v2)

While the rfkill class does work with just get_state(), it doesn't work
well on devices that are subject to external events that cause rfkill state
changes.

Document that rfkill_force_state() is required in those cases.

Signed-off-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoath5k: use positive logic for HP laptop LEDs
Bob Copeland [Tue, 15 Jul 2008 17:07:16 +0000]
ath5k: use positive logic for HP laptop LEDs

Helge Deller reports that HP laptops (NC4010 and NC6000) use active-
high signals to turn on the LEDs.  Previous code used active-low for
all devices.

Signed-off-by: Bob Copeland <me@bobcopeland.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoAth5k: suspend/resume fixes
Jiri Slaby [Tue, 15 Jul 2008 15:44:21 +0000]
Ath5k: suspend/resume fixes

- free and re-request irq since it might have changed during suspend
- disable and enable msi
- don't set D0 state of the device, it's already done by the PCI layer
- do restore_state before enable_device, it's safer
- check ath5k_init return value

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Cc: Luis R. Rodriguez <mcgrof@gmail.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoAth5k: fix dma operation
Jiri Slaby [Tue, 15 Jul 2008 15:44:43 +0000]
Ath5k: fix dma operation

Don't sync
- coherent mapping (descriptors)
- before unmap, it's useless
- (wrongly anyway -- for_cpu) beacon skb, it's just mapped,
  so by the device yet

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Cc: Luis R. Rodriguez <mcgrof@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoAth5k: flush work
Jiri Slaby [Tue, 15 Jul 2008 15:44:20 +0000]
Ath5k: flush work

Make sure that the irq is not in progress after stop. This means
two things:
- ensure the intr setting register is set by flushing posted values
- call synchronize_irq() after that

Also flush stop tx write, inform callers of the tx stop about still
pending transfers (unsuccessful stop) and finally don't wait another
3ms in ath5k_rx_stop, since ath5k_hw_stop_rx_dma ensures transfer to
be finished.

Make sure all writes will be ordered in respect to locks by mmiowb().

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Cc: Luis R. Rodriguez <mcgrof@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoAth5k: kill tasklets on shutdown
Jiri Slaby [Tue, 15 Jul 2008 15:44:19 +0000]
Ath5k: kill tasklets on shutdown

Don't forget to kill tasklets on stop to not panic if they
fire after freeing some structures.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Cc: Luis R. Rodriguez <mcgrof@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoAth5k: fix memory corruption
Jiri Slaby [Tue, 15 Jul 2008 15:44:18 +0000]
Ath5k: fix memory corruption

When signal is noisy, hardware can use all RX buffers and since the last
entry in the list is self-linked, it overwrites the entry until we link
new buffers.

Ensure that we don't free this last one until we are 100% sure that it
is not used by the hardware anymore to not cause memory curruption as
can be seen below.

This is done by checking next buffer in the list. Even after that we
know that the hardware refetched the new link and proceeded further
(the next buffer is ready) we can finally free the overwritten buffer.

We discard it since the status in its descriptor is overwritten (OR-ed
by new status) too.

=============================================================================
BUG kmalloc-4096: Poison overwritten
-----------------------------------------------------------------------------

INFO: 0xffff810067419060-0xffff810067419667. First byte 0x8 instead of 0x6b
INFO: Allocated in dev_alloc_skb+0x18/0x30 age=1118 cpu=1 pid=0
INFO: Freed in skb_release_data+0x85/0xd0 age=1105 cpu=1 pid=3718
INFO: Slab 0xffffe200019d0600 objects=7 used=0 fp=0xffff810067419048 flags=0x40000000000020c3
INFO: Object 0xffff810067419048 @offset=4168 fp=0xffff81006741c120

Bytes b4 0xffff810067419038:  4f 0b 02 00 01 00 00 00 5a 5a 5a 5a 5a 5a 5a 5a O.......ZZZZZZZZ
  Object 0xffff810067419048:  6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkkkkkkkkkkkkkk
  Object 0xffff810067419058:  6b 6b 6b 6b 6b 6b 6b 6b 08 42 30 00 00 0b 6b 80 kkkkkkkk.B0...k.
  Object 0xffff810067419068:  f0 5d 00 4f 62 08 a3 64 00 0c 42 16 52 e4 f0 5a 360].Ob.243d..B.R344360Z
  Object 0xffff810067419078:  68 81 00 00 7b a5 b4 be 7d 3b 8f 53 cd d5 de 12 h...{245264276};.S315325336.
  Object 0xffff810067419088:  96 10 0b 89 48 54 23 41 0f 4e 2d b9 37 c3 cb 29 ....HT#A.N-2717303313)
  Object 0xffff810067419098:  d1 e0 de 14 8a 57 2a cc 3b 44 0d 78 7a 19 12 15 321340336..W*314;D.xz...
  Object 0xffff8100674190a8:  a9 ec d4 35 a8 10 ec 8c 40 a7 06 0a 51 a7 48 bb 2513543245250.354.@247..Q247H273
  Object 0xffff8100674190b8:  3e cf a1 c7 38 60 63 3f 51 15 c7 20 eb ba 65 30 >ϡ3078`c?Q.307.353272e0
 Redzone 0xffff81006741a048:  bb bb bb bb bb bb bb bb                         273273273273273273273273
 Padding 0xffff81006741a088:  5a 5a 5a 5a 5a 5a 5a 5a                         ZZZZZZZZ
Pid: 3297, comm: ath5k_pci Not tainted 2.6.26-rc8-mm1_64 #427

Call Trace:
 [<ffffffff802a7306>] print_trailer+0xf6/0x150
 [<ffffffff802a7485>] check_bytes_and_report+0x125/0x180
 [<ffffffff802a75dc>] check_object+0xac/0x260
 [<ffffffff802a9308>] __slab_alloc+0x368/0x6d0
 [<ffffffff80544f82>] ? wireless_send_event+0x142/0x310
 [<ffffffff804b1bd4>] ? __alloc_skb+0x44/0x150
 [<ffffffff80544f82>] ? wireless_send_event+0x142/0x310
 [<ffffffff802aa853>] __kmalloc_track_caller+0xc3/0xf0
 [<ffffffff804b1bfe>] __alloc_skb+0x6e/0x150
[... stack snipped]

FIX kmalloc-4096: Restoring 0xffff810067419060-0xffff810067419667=0x6b

FIX kmalloc-4096: Marking all objects used

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Cc: Luis R. Rodriguez <mcgrof@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agomac80211: remove IEEE80211_HW_HOST_GEN_BEACON_TEMPLATE flag
Johannes Berg [Tue, 15 Jul 2008 08:10:01 +0000]
mac80211: remove IEEE80211_HW_HOST_GEN_BEACON_TEMPLATE flag

I forgot this in the previous patch that made it unused.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

11 years agoLinux 2.6.27-rc1
Linus Torvalds [Tue, 29 Jul 2008 02:40:31 +0000]
Linux 2.6.27-rc1

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus
Linus Torvalds [Tue, 29 Jul 2008 01:16:26 +0000]
Merge git://git./linux/kernel/git/rusty/linux-2.6-for-linus

* git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus:
  lguest: turn Waker into a thread, not a process
  lguest: Enlarge virtio rings
  lguest: Use GSO/IFF_VNET_HDR extensions on tun/tap
  lguest: Remove 'network: no dma buffer!' warning
  lguest: Adaptive timeout
  lguest: Tell Guest net not to notify us on every packet xmit
  lguest: net block unneeded receive queue update notifications
  lguest: wrap last_avail accesses.
  lguest: use cpu capability accessors
  lguest: virtio-rng support
  lguest: Support assigning a MAC address
  lguest: Don't leak /dev/zero fd
  lguest: fix verbose printing of device features.
  lguest: fix switcher_page leak on unload
  lguest: Guest int3 fix
  lguest: set max_pfn_mapped, growl loudly at Yinghai Lu

11 years agoMerge branch 'for-linus' of git://git.o-hand.com/linux-mfd
Linus Torvalds [Tue, 29 Jul 2008 01:15:41 +0000]
Merge branch 'for-linus' of git://git.o-hand.com/linux-mfd

* 'for-linus' of git://git.o-hand.com/linux-mfd:
  mfd: accept pure device as a parent, not only platform_device
  mfd: add platform_data to mfd_cell
  mfd: Coding style fixes
  mfd: Use to_platform_device instead of container_of

11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes...
Linus Torvalds [Tue, 29 Jul 2008 01:14:24 +0000]
Merge branch 'for-linus' of git://git./linux/kernel/git/jbarnes/pci-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes/pci-2.6: (21 commits)
  x86/PCI: use dev_printk when possible
  PCI: add D3 power state avoidance quirk
  PCI: fix bogus "'device' may be used uninitialized" warning in pci_slot
  PCI: add an option to allow ASPM enabled forcibly
  PCI: disable ASPM on pre-1.1 PCIe devices
  PCI: disable ASPM per ACPI FADT setting
  PCI MSI: Don't disable MSIs if the mask bit isn't supported
  PCI: handle 64-bit resources better on 32-bit machines
  PCI: rewrite PCI BAR reading code
  PCI: document pci_target_state
  PCI hotplug: fix typo in pcie hotplug output
  x86 gart: replace to_pages macro with iommu_num_pages
  x86, AMD IOMMU: replace to_pages macro with iommu_num_pages
  iommu: add iommu_num_pages helper function
  dma-coherent: add documentation to new interfaces
  Cris: convert to using generic dma-coherent mem allocator
  Sh: use generic per-device coherent dma allocator
  ARM: support generic per-device coherent dma mem
  Generic dma-coherent: fix DMA_MEMORY_EXCLUSIVE
  x86: use generic per-device dma coherent allocator
  ...

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
Linus Torvalds [Tue, 29 Jul 2008 01:13:48 +0000]
Merge git://git./linux/kernel/git/jejb/scsi-rc-fixes-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6:
  [SCSI] qla2xxx: fix msleep compile error

11 years agoFix 'get_user_pages_fast()' with non-page-aligned start address
Linus Torvalds [Tue, 29 Jul 2008 00:54:21 +0000]
Fix 'get_user_pages_fast()' with non-page-aligned start address

Alexey Dobriyan reported trouble with LTP with the new fast-gup code,
and Johannes Weiner debugged it to non-page-aligned addresses, where the
new get_user_pages_fast() code would do all the wrong things, including
just traversing past the end of the requested area due to 'addr' never
matching 'end' exactly.

This is not a pretty fix, and we may actually want to move the alignment
into generic code, leaving just the core code per-arch, but Alexey
verified that the vmsplice01 LTP test doesn't crash with this.

Reported-and-tested-by: Alexey Dobriyan <adobriyan@gmail.com>
Debugged-by: Johannes Weiner <hannes@saeurebad.de>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agolguest: turn Waker into a thread, not a process
Rusty Russell [Tue, 29 Jul 2008 14:58:38 +0000]
lguest: turn Waker into a thread, not a process

lguest uses a Waker process to break it out of the kernel (ie.
actually running the guest) when file descriptor needs attention.

Changing this from a process to a thread somewhat simplifies things:
it can directly access the fd_set of things to watch.  More
importantly, it means that the Waker can see Guest memory correctly,
so /dev/vring file descriptors will work as anticipated (the
alternative is to actually mmap MAP_SHARED, but you can't do that with
/dev/zero).

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Enlarge virtio rings
Rusty Russell [Tue, 29 Jul 2008 14:58:37 +0000]
lguest: Enlarge virtio rings

With big packets, 128 entries is a little small.

Guest -> Host 1GB TCP:
Before: 8.43625 seconds xmit 95640 recv 198266 timeout 49771 usec 1252
After: 8.01099 seconds xmit 49200 recv 102263 timeout 26014 usec 2118

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Use GSO/IFF_VNET_HDR extensions on tun/tap
Rusty Russell [Tue, 29 Jul 2008 14:58:37 +0000]
lguest: Use GSO/IFF_VNET_HDR extensions on tun/tap

Guest -> Host 1GB TCP:
Before 20.1974 seconds xmit 214510 recv 5 timeout 214491 usec 278
After 8.43625 seconds xmit 95640 recv 198266 timeout 49771 usec 1252

Host -> Guest 1GB TCP:
Before: Seconds 9.98854 xmit 172166 recv 5344 timeout 172157 usec 251
After: Seconds 5.72803 xmit 244322 recv 9919 timeout 244302 usec 156

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Remove 'network: no dma buffer!' warning
Rusty Russell [Tue, 29 Jul 2008 14:58:36 +0000]
lguest: Remove 'network: no dma buffer!' warning

This warning can happen a lot under load, and it should be warnx not
warn anwyay.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Adaptive timeout
Rusty Russell [Tue, 29 Jul 2008 14:58:36 +0000]
lguest: Adaptive timeout

Since the correct timeout value varies, use a heuristic which adjusts
the timeout depending on how many packets we've seen.  This gives
slightly worse results, but doesn't need tweaking when GSO is
introduced.

500 usec 19.1887 xmit 561141 recv 1 timeout 559657
Dynamic (278) 20.1974 xmit 214510 recv 5 timeout 214491 usec 278

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Tell Guest net not to notify us on every packet xmit
Rusty Russell [Tue, 29 Jul 2008 14:58:35 +0000]
lguest: Tell Guest net not to notify us on every packet xmit

virtio_ring has the ability to suppress notifications.  This prevents
a guest exit for every packet, but we need to set a timer on packet
receipt to re-check if there were any remaining packets.

Here are the times for 1G TCP Guest->Host with different timeout
settings (it matters because the TCP window doesn't grow big enough to
fill the entire buffer):

Timeout value Seconds Xmit/Recv/Timeout
None (before) 25.3784 xmit 7750233 recv 1
2500 usec 62.5119 xmit 207020 recv 2 timeout 207020
1000 usec 34.5379 xmit 207003 recv 2 timeout 207003
750 usec 29.2305 xmit 207002 recv 1 timeout 207002
500 usec 19.1887 xmit 561141 recv 1 timeout 559657
250 usec 20.0465 xmit 214128 recv 2 timeout 214110
100 usec 19.2583 xmit 561621 recv 1 timeout 560153

(Note that these values are sensitive to the GSO patches which come
 later, and probably other traffic-related variables, so take with a
 large grain of salt).

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: net block unneeded receive queue update notifications
Rusty Russell [Tue, 29 Jul 2008 14:58:35 +0000]
lguest: net block unneeded receive queue update notifications

Number of exits transmitting 10GB Guest->Host before:
network xmit 7858610 recv 118136

After:
network xmit 7750233 recv 1

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: wrap last_avail accesses.
Rusty Russell [Tue, 29 Jul 2008 14:58:34 +0000]
lguest: wrap last_avail accesses.

To simplify the transition to when we publish indices in the ring
(and make shuffling my patch queue easier), wrap them in a lg_last_avail()
macro.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: use cpu capability accessors
Andrew Morton [Mon, 9 Jun 2008 23:22:48 +0000]
lguest: use cpu capability accessors

To support my little make-x86-bitops-use-proper-typechecking projectlet.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: virtio-rng support
Rusty Russell [Tue, 29 Jul 2008 14:58:33 +0000]
lguest: virtio-rng support

This is a simple patch to add support for the virtio "hardware random
generator" to lguest.  It gets about 1.2 MB/sec reading from /dev/hwrng
in the guest.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Support assigning a MAC address
Mark McLoughlin [Tue, 29 Jul 2008 14:58:33 +0000]
lguest: Support assigning a MAC address

If you've got a nice DHCP configuration which maps MAC
addresses to specific IP addresses, then you're going to
want to start your guest with one of those MAC addresses.

Also, in Fedora, we have persistent network interface naming
based on the MAC address, so with randomly assigned
addresses you're soon going to hit eth13. Who knows what
will happen then!

Allow assigning a MAC address to the network interface with
e.g.

  --tunnet=bridge:eth0:00:FF:95:6B:DA:3D

or:

  --tunnet=192.168.121.1:00:FF:95:6B:DA:3D

which is pretty unintelligable, but ...

(includes Rusty's minor rework)

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Don't leak /dev/zero fd
Mark McLoughlin [Fri, 13 Jun 2008 13:04:58 +0000]
lguest: Don't leak /dev/zero fd

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: fix verbose printing of device features.
Rusty Russell [Tue, 29 Jul 2008 14:58:32 +0000]
lguest: fix verbose printing of device features.

%02x is more appropriate for bytes than %08x.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: fix switcher_page leak on unload
Johannes Weiner [Tue, 8 Jul 2008 08:29:42 +0000]
lguest: fix switcher_page leak on unload

map_switcher allocates the array, unmap_switcher has to free it
accordingly.

Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: Guest int3 fix
Rusty Russell [Tue, 29 Jul 2008 14:58:31 +0000]
lguest: Guest int3 fix

Ron Minnich noticed that guest userspace gets a GPF when it tries to int3:
we need to copy the privilege level from the guest-supplied IDT to the real
IDT.  int3 is the only common case where guest userspace expects to invoke
an interrupt, so that's the symptom of failing to do this.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

11 years agolguest: set max_pfn_mapped, growl loudly at Yinghai Lu
Rusty Russell [Tue, 29 Jul 2008 14:58:29 +0000]
lguest: set max_pfn_mapped, growl loudly at Yinghai Lu

6af61a7614a306fe882a0c2b4ddc63b65aa66efc 'x86: clean up max_pfn_mapped
usage - 32-bit' makes the following comment:

    XEN PV and lguest may need to assign max_pfn_mapped too.

But no CC.  Yinghai, wasting fellow developers' time is a VERY bad
habit.  If you do it again, I will hunt you down and try to extract
the three hours of my life I just lost :)

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>

11 years agomfd: accept pure device as a parent, not only platform_device
Dmitry Baryshkov [Mon, 28 Jul 2008 23:30:26 +0000]
mfd: accept pure device as a parent, not only platform_device

Signed-off-by: Dmitry Baryshkov <dbaryshkov@gmail.com>
Signed-off-by: Samuel Ortiz <sameo@openedhand.com>

11 years agoinclude/asm-generic/pgtable-nopmd.h: macros are noxious, reason #435
Andrew Morton [Mon, 28 Jul 2008 22:46:40 +0000]
include/asm-generic/pgtable-nopmd.h: macros are noxious, reason #435

arch/x86/mm/pgtable.c: In function 'pgd_mop_up_pmds':
  arch/x86/mm/pgtable.c:194: warning: unused variable 'pmd'

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agosh7760fb: write colormap value to hardware
Manuel Lauss [Mon, 28 Jul 2008 22:46:39 +0000]
sh7760fb: write colormap value to hardware

The computed color value is never actually written to hardware
colormap register.

Signed-off-by: Manuel Lauss <mano@roarinelk.homelinux.net>
Cc: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
Cc: Munakata Hisao <munakata.hisao@renesas.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoeCryptfs: use page_alloc not kmalloc to get a page of memory
Eric Sandeen [Mon, 28 Jul 2008 22:46:39 +0000]
eCryptfs: use page_alloc not kmalloc to get a page of memory

With SLUB debugging turned on in 2.6.26, I was getting memory corruption
when testing eCryptfs.  The root cause turned out to be that eCryptfs was
doing kmalloc(PAGE_CACHE_SIZE); virt_to_page() and treating that as a nice
page-aligned chunk of memory.  But at least with SLUB debugging on, this
is not always true, and the page we get from virt_to_page does not
necessarily match the PAGE_CACHE_SIZE worth of memory we got from kmalloc.

My simple testcase was 2 loops doing "rm -f fileX; cp /tmp/fileX ." for 2
different multi-megabyte files.  With this change I no longer see the
corruption.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Acked-by: Michael Halcrow <mhalcrow@us.ibm.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agogpio: fix build on CONFIG_GPIO_SYSFS=n
Atsushi Nemoto [Mon, 28 Jul 2008 22:46:38 +0000]
gpio: fix build on CONFIG_GPIO_SYSFS=n

If CONFIG_GENERIC_GPIO=y && CONFIG_GPIO_SYSFS=n, gpio_export() in
asm-generic/gpio.h refers -ENOSYS and causes build error.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agobio-integrity: remove EXPORT_SYMBOL for bio_integrity_init_slab()
Yoichi Yuasa [Mon, 28 Jul 2008 22:46:37 +0000]
bio-integrity: remove EXPORT_SYMBOL for bio_integrity_init_slab()

I got section mismatch message about bio_integrity_init_slab().

WARNING: fs/built-in.o(__ksymtab+0xb60): Section mismatch in reference from the variable __ksymtab_bio_integrity_init_slab to the function .init.text:bio_integrity_init_slab()

The symbol bio_integrity_init_slab is exported and annotated __init Fix
this by removing the __init annotation of bio_integrity_init_slab or drop
the export.

It only call from init_bio().  The EXPORT_SYMBOL() can be removed.

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agovfs: pagecache usage optimization for pagesize!=blocksize
Hisashi Hifumi [Mon, 28 Jul 2008 22:46:36 +0000]
vfs: pagecache usage optimization for pagesize!=blocksize

When we read some part of a file through pagecache, if there is a
pagecache of corresponding index but this page is not uptodate, read IO
is issued and this page will be uptodate.

I think this is good for pagesize == blocksize environment but there is
room for improvement on pagesize != blocksize environment.  Because in
this case a page can have multiple buffers and even if a page is not
uptodate, some buffers can be uptodate.

So I suggest that when all buffers which correspond to a part of a file
that we want to read are uptodate, use this pagecache and copy data from
this pagecache to user buffer even if a page is not uptodate.  This can
reduce read IO and improve system throughput.

I wrote a benchmark program and got result number with this program.

This benchmark do:

  1: mount and open a test file.

  2: create a 512MB file.

  3: close a file and umount.

  4: mount and again open a test file.

  5: pwrite randomly 300000 times on a test file.  offset is aligned
     by IO size(1024bytes).

  6: measure time of preading randomly 100000 times on a test file.

The result was:
2.6.26
        330 sec

2.6.26-patched
        226 sec

Arch:i386
Filesystem:ext3
Blocksize:1024 bytes
Memory: 1GB

On ext3/4, a file is written through buffer/block.  So random read/write
mixed workloads or random read after random write workloads are optimized
with this patch under pagesize != blocksize environment.  This test result
showed this.

The benchmark program is as follows:

#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <time.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mount.h>

#define LEN 1024
#define LOOP 1024*512 /* 512MB */

main(void)
{
unsigned long i, offset, filesize;
int fd;
char buf[LEN];
time_t t1, t2;

if (mount("/dev/sda1", "/root/test1/", "ext3", 0, 0) < 0) {
perror("cannot mount\n");
exit(1);
}
memset(buf, 0, LEN);
fd = open("/root/test1/testfile", O_CREAT|O_RDWR|O_TRUNC);
if (fd < 0) {
perror("cannot open file\n");
exit(1);
}
for (i = 0; i < LOOP; i++)
write(fd, buf, LEN);
close(fd);
if (umount("/root/test1/") < 0) {
perror("cannot umount\n");
exit(1);
}
if (mount("/dev/sda1", "/root/test1/", "ext3", 0, 0) < 0) {
perror("cannot mount\n");
exit(1);
}
fd = open("/root/test1/testfile", O_RDWR);
if (fd < 0) {
perror("cannot open file\n");
exit(1);
}

filesize = LEN * LOOP;
for (i = 0; i < 300000; i++){
offset = (random() % filesize) & (~(LEN - 1));
pwrite(fd, buf, LEN, offset);
}
printf("start test\n");
time(&t1);
for (i = 0; i < 100000; i++){
offset = (random() % filesize) & (~(LEN - 1));
pread(fd, buf, LEN, offset);
}
time(&t2);
printf("%ld sec\n", t2-t1);
close(fd);
if (umount("/root/test1/") < 0) {
perror("cannot umount\n");
exit(1);
}
}

Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Jan Kara <jack@ucw.cz>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agokdump: update kdump documentation as kexec-tools-resting has been renamed kexec-tools
Simon Horman [Mon, 28 Jul 2008 22:46:34 +0000]
kdump: update kdump documentation as kexec-tools-resting has been renamed kexec-tools

Signed-off-by: Simon Horman <horms@verge.net.au>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agospi_s3c24xx: really assign busnum
Ben Dooks [Mon, 28 Jul 2008 22:46:33 +0000]
spi_s3c24xx: really assign busnum

The original "Pass the bus number we expect the S3C24XX SPI driver to
attach to via the platform data." [1] patch was mis-sent, and missed two
important parts of the diff, which was to actually set the bus_num field
and add the relevant field to the platform data.

The previous commit 50f426b55d919dd017af35bb6a08753d1f262920 promised to
add a bus_num field, but failed to include the two hunks that added this
field to include/asm-arm/arch-s3c2410/spi.h and then pass it to the spi
core when creating the new master field in drivers/spi/spi_s3c24xx.c.

[1] git commit 50f426b55d919dd017af35bb6a08753d1f262920

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agompc52xx_psc_spi: fix block transfer
Luotao Fu [Mon, 28 Jul 2008 22:46:32 +0000]
mpc52xx_psc_spi: fix block transfer

The block transfer routine in the mpc52xx psc spi driver misinterpret
the datasheet.  According to the processor datasheet the chipselect is
held as long as the EOF is not written.

Theoretically blocks of any sizes can be transferred in this way.  The
old routine however writes an EOF after every word, which has the size
of size_of_word.  This makes the transfer slow.

Also fixed some duplicate code.

Signed-off-by: Luotao Fu <l.fu@pengutronix.de>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agomm/hugetlb.c must #include <asm/io.h>
Adrian Bunk [Mon, 28 Jul 2008 22:46:30 +0000]
mm/hugetlb.c must #include <asm/io.h>

This patch fixes the following build error on sh caused by commit
aa888a74977a8f2120ae9332376e179c39a6b07d ("hugetlb: support larger than
MAX_ORDER"):

  mm/hugetlb.c: In function 'alloc_bootmem_huge_page':
  mm/hugetlb.c:958: error: implicit declaration of function 'virt_to_phys'

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agommu-notifiers: core
Andrea Arcangeli [Mon, 28 Jul 2008 22:46:29 +0000]
mmu-notifiers: core

With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages.
 There are secondary MMUs (with secondary sptes and secondary tlbs) too.
sptes in the kvm case are shadow pagetables, but when I say spte in
mmu-notifier context, I mean "secondary pte".  In GRU case there's no
actual secondary pte and there's only a secondary tlb because the GRU
secondary MMU has no knowledge about sptes and every secondary tlb miss
event in the MMU always generates a page fault that has to be resolved by
the CPU (this is not the case of KVM where the a secondary tlb miss will
walk sptes in hardware and it will refill the secondary tlb transparently
to software if the corresponding spte is present).  The same way
zap_page_range has to invalidate the pte before freeing the page, the spte
(and secondary tlb) must also be invalidated before any page is freed and
reused.

Currently we take a page_count pin on every page mapped by sptes, but that
means the pages can't be swapped whenever they're mapped by any spte
because they're part of the guest working set.  Furthermore a spte unmap
event can immediately lead to a page to be freed when the pin is released
(so requiring the same complex and relatively slow tlb_gather smp safe
logic we have in zap_page_range and that can be avoided completely if the
spte unmap event doesn't require an unpin of the page previously mapped in
the secondary MMU).

The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know
when the VM is swapping or freeing or doing anything on the primary MMU so
that the secondary MMU code can drop sptes before the pages are freed,
avoiding all page pinning and allowing 100% reliable swapping of guest
physical address space.  Furthermore it avoids the code that teardown the
mappings of the secondary MMU, to implement a logic like tlb_gather in
zap_page_range that would require many IPI to flush other cpu tlbs, for
each fixed number of spte unmapped.

To make an example: if what happens on the primary MMU is a protection
downgrade (from writeable to wrprotect) the secondary MMU mappings will be
invalidated, and the next secondary-mmu-page-fault will call
get_user_pages and trigger a do_wp_page through get_user_pages if it
called get_user_pages with write=1, and it'll re-establishing an updated
spte or secondary-tlb-mapping on the copied page.  Or it will setup a
readonly spte or readonly tlb mapping if it's a guest-read, if it calls
get_user_pages with write=0.  This is just an example.

This allows to map any page pointed by any pte (and in turn visible in the
primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an
full MMU with both sptes and secondary-tlb like the shadow-pagetable layer
with kvm), or a remote DMA in software like XPMEM (hence needing of
schedule in XPMEM code to send the invalidate to the remote node, while no
need to schedule in kvm/gru as it's an immediate event like invalidating
primary-mmu pte).

At least for KVM without this patch it's impossible to swap guests
reliably.  And having this feature and removing the page pin allows
several other optimizations that simplify life considerably.

Dependencies:

1) mm_take_all_locks() to register the mmu notifier when the whole VM
   isn't doing anything with "mm".  This allows mmu notifier users to keep
   track if the VM is in the middle of the invalidate_range_begin/end
   critical section with an atomic counter incraese in range_begin and
   decreased in range_end.  No secondary MMU page fault is allowed to map
   any spte or secondary tlb reference, while the VM is in the middle of
   range_begin/end as any page returned by get_user_pages in that critical
   section could later immediately be freed without any further
   ->invalidate_page notification (invalidate_range_begin/end works on
   ranges and ->invalidate_page isn't called immediately before freeing
   the page).  To stop all page freeing and pagetable overwrites the
   mmap_sem must be taken in write mode and all other anon_vma/i_mmap
   locks must be taken too.

2) It'd be a waste to add branches in the VM if nobody could possibly
   run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if
   CONFIG_KVM=m/y.  In the current kernel kvm won't yet take advantage of
   mmu notifiers, but this already allows to compile a KVM external module
   against a kernel with mmu notifiers enabled and from the next pull from
   kvm.git we'll start using them.  And GRU/XPMEM will also be able to
   continue the development by enabling KVM=m in their config, until they
   submit all GRU/XPMEM GPLv2 code to the mainline kernel.  Then they can
   also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n).
   This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM
   are all =n.

The mmu_notifier_register call can fail because mm_take_all_locks may be
interrupted by a signal and return -EINTR.  Because mmu_notifier_reigster
is used when a driver startup, a failure can be gracefully handled.  Here
an example of the change applied to kvm to register the mmu notifiers.
Usually when a driver startups other allocations are required anyway and
-ENOMEM failure paths exists already.

 struct  kvm *kvm_arch_create_vm(void)
 {
        struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
+       int err;

        if (!kvm)
                return ERR_PTR(-ENOMEM);

        INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);

+       kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops;
+       err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm);
+       if (err) {
+               kfree(kvm);
+               return ERR_PTR(err);
+       }
+
        return kvm;
 }

mmu_notifier_unregister returns void and it's reliable.

The patch also adds a few needed but missing includes that would prevent
kernel to compile after these changes on non-x86 archs (x86 didn't need
them by luck).

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix mm/filemap_xip.c build]
[akpm@linux-foundation.org: fix mm/mmu_notifier.c build]
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Kanoj Sarcar <kanojsarcar@yahoo.com>
Cc: Roland Dreier <rdreier@cisco.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Avi Kivity <avi@qumranet.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Chris Wright <chrisw@redhat.com>
Cc: Marcelo Tosatti <marcelo@kvack.org>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Izik Eidus <izike@qumranet.com>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agommu-notifiers: add mm_take_all_locks() operation
Andrea Arcangeli [Mon, 28 Jul 2008 22:46:26 +0000]
mmu-notifiers: add mm_take_all_locks() operation

mm_take_all_locks holds off reclaim from an entire mm_struct.  This allows
mmu notifiers to register into the mm at any time with the guarantee that
no mmu operation is in progress on the mm.

This operation locks against the VM for all pte/vma/mm related operations
that could ever happen on a certain mm.  This includes vmtruncate,
try_to_unmap, and all page faults.

The caller must take the mmap_sem in write mode before calling
mm_take_all_locks().  The caller isn't allowed to release the mmap_sem
until mm_drop_all_locks() returns.

mmap_sem in write mode is required in order to block all operations that
could modify pagetables and free pages without need of altering the vma
layout (for example populate_range() with nonlinear vmas).  It's also
needed in write mode to avoid new anon_vmas to be associated with existing
vmas.

A single task can't take more than one mm_take_all_locks() in a row or it
would deadlock.

mm_take_all_locks() and mm_drop_all_locks are expensive operations that
may have to take thousand of locks.

mm_take_all_locks() can fail if it's interrupted by signals.

When mmu_notifier_register returns, we must be sure that the driver is
notified if some task is in the middle of a vmtruncate for the 'mm' where
the mmu notifier was registered (mmu_notifier_invalidate_range_start/end
is run around the vmtruncation but mmu_notifier_register can run after
mmu_notifier_invalidate_range_start and before
mmu_notifier_invalidate_range_end).  Same problem for rmap paths.  And
we've to remove page pinning to avoid replicating the tlb_gather logic
inside KVM (and GRU doesn't work well with page pinning regardless of
needing tlb_gather), so without mm_take_all_locks when vmtruncate frees
the page, kvm would have no way to notice that it mapped into sptes a page
that is going into the freelist without a chance of any further
mmu_notifier notification.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Kanoj Sarcar <kanojsarcar@yahoo.com>
Cc: Roland Dreier <rdreier@cisco.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Avi Kivity <avi@qumranet.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Chris Wright <chrisw@redhat.com>
Cc: Marcelo Tosatti <marcelo@kvack.org>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Izik Eidus <izike@qumranet.com>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>