Fix ABBA deadlock between lock_cpu_hotplug() and the cpuset
callback_mutex lock.
It only happens on cpu_exclusive cpusets, due to the dynamic
sched domain code trying to take the cpu hotplug lock inside
the cpuset callback_mutex lock.
This bug has apparently been here for several months, but didn't
get hit until the right customer load on a large system.
This fix appears right from inspection, but it will take a few
more days running it on that customers workload to be confident
we nailed it. We don't have any other reproducible test case.
The cpu_hotplug_lock() tends to cover large runs of code.
The other places that hold both that lock and the cpuset callback
mutex lock always nest the cpuset lock inside the hotplug lock.
This place tries to do the reverse, risking an ABBA deadlock.
This is in the cpuset_rmdir() code, where we:
* take the callback_mutex lock
* mark the cpuset CS_REMOVED
* call update_cpu_domains for cpu_exclusive cpusets
* in that call, take the cpu_hotplug lock if the
cpuset is marked for removal.
Thanks to Jack Steiner for identifying this deadlock.
The fix is to tear down the dynamic sched domain before we grab
the cpuset callback_mutex lock. This way, the two locks are
serialized, with the hotplug lock taken and released before
trying for the cpuset lock.
I suspect that this bug was introduced when I changed the
cpuset locking from one lock to two. The dynamic sched domain
dependency on cpu_exclusive cpusets and its hotplug hooks were
added to this code earlier, when cpusets had only a single lock.
It may well have been fine then.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The CPU hotplug locking was quite messy, with a recursive lock to
handle the fact that both the actual up/down sequence wanted to
protect itself from being re-entered, but the callbacks that it
called also tended to want to protect themselves from CPU events.
This splits the lock into two (one to serialize the whole hotplug
sequence, the other to protect against the CPU present bitmaps
changing). The latter still allows recursive usage because some
subsystems (ondemand policy for cpufreq at least) had already gotten
too used to the lax locking, but the locking mistakes are hopefully
now less fundamental, and we now warn about recursive lock usage
when we see it, in the hope that it can be fixed.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Shutting down the ondemand policy was fraught with potential
problems, causing issues for SMP suspend (which wants to hot-
unplug) all but the last CPU.
This should fix at least the worst problems (divide-by-zero
and infinite wait for the workqueue to shut down).
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6: (21 commits)
[TIPC]: Removing useless casts
[IPV4]: Fix nexthop realm dumping for multipath routes
[DUMMY]: Avoid an oops when dummy_init_one() failed
[IFB] After ifb_init_one() failed, i is increased. Decrease
[NET]: Fix reversed error test in netif_tx_trylock
[MAINTAINERS]: Mark LAPB as Oprhan.
[NET]: Conversions from kmalloc+memset to k(z|c)alloc.
[NET]: sun happymeal, little pci cleanup
[IrDA]: Use alloc_skb() in IrDA TX path
[I/OAT]: Remove pci_module_init() from Intel I/OAT DMA engine
[I/OAT]: net/core/user_dma.c should #include <net/netdma.h>
[SCTP]: ADDIP: Don't use an address as source until it is ASCONF-ACKed
[SCTP]: Set chunk->data_accepted only if we are going to accept it.
[SCTP]: Verify all the paths to a peer via heartbeat before using them.
[SCTP]: Unhash the endpoint in sctp_endpoint_free().
[SCTP]: Check for NULL arg to sctp_bucket_destroy().
[PKT_SCHED] netem: Fix slab corruption with netem (2nd try)
[WAN]: Converted synclink drivers to use netif_carrier_*()
[WAN]: Cosmetic changes to N2 and C101 drivers
[WAN]: Added missing netif_dormant_off() to generic HDLC
...
Routing realms exist per nexthop, but are only returned to userspace
for the first nexthop. This is due to the fact that iproute2 only
allows to set the realm for the first nexthop and the kernel refuses
multipath routes where only a single realm is present.
Dump all realms for multipath routes to enable iproute to correctly
display them.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
It before entering in the loop for freeing the other ifb devices.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
A non-zero return value indicates success from spin_trylock,
not error.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use pci_register_driver instead of pci_module_init. Use PCI_DEVICE macro.
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As pointed out by Christoph Hellwig, dev_alloc_skb() is not intended to be
used for allocating TX sk_buff. The IrDA stack was exclusively calling
dev_alloc_skb() on the TX path, and this patch fixes that.
Signed-off-by: Samuel Ortiz <samuel@sortiz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Changes pci_module_init() to pci_register_driver().
Signed-off-by: Henrik Kretzschmar <henne@nachtwindheim.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Every file should #include the headers containing the prototypes for
its global functions.
Especially in cases like this one where gcc can tell us through a
compile error that the prototype was wrong...
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This implements Rules D1 and D4 of Sec 4.3 in the ADDIP draft.
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently there is a code path in sctp_eat_data() where it is possible
to set this flag even when we are dropping this chunk.
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements Path Initialization procedure as described in
Sec 2.36 of RFC4460.
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This prevents a race between the close of a socket and receive of an
incoming packet.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
CONFIG_DEBUG_SLAB found the following bug:
netem_enqueue() in sch_netem.c gets a pointer inside a slab object:
struct netem_skb_cb *cb = (struct netem_skb_cb *)skb->cb;
But then, the slab object may be freed:
skb = skb_unshare(skb, GFP_ATOMIC)
cb is still pointing inside the freed skb, so here is a patch to
initialize cb later, and make it clear that initializing it sooner
is a bad idea.
[From Stephen Hemminger: leave cb unitialized in order to let gcc
complain in case of use before initialization]
Signed-off-by: Guillaume Chazarain <guichaz@yahoo.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
WAN: Converted synclink drivers to use netif_carrier_*() instead
of hdlc_set_carrier().
Signed-off-by: Krzysztof Halasa <khc@pm.waw.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
WAN: Fixed a problem with PPP/raw HDLC/X.25 protocols not doing
netif_dormant_off() at startup.
Signed-off-by: Krzysztof Halasa <khc@pm.waw.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we always zero the IPCB->opts in ip_rcv, it is no longer
necessary to do so before calling netif_rx for tunneled packets.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Just do a linear enumeration so that we handle sun4d systems
correctly. As a consequence, eliminate the hard coded keyboard and
mouse channel line values, use the CONS_{KEYB,MS} flags instead.
Also, report the keyboard/mouse Zilog channels just like the uart ones
do.
Signed-off-by: David S. Miller <davem@davemloft.net>
By using for_each_node_by_type().
Also, correct a spurioud test in check_cpu_node() on sparc64.
It is only called with nodes that have device_type "cpu".
Signed-off-by: David S. Miller <davem@davemloft.net>
1) Explicitly traverse to the root looking for the "sbi".
2) Grab the "board#" property from the sbi's parent and
verify that this parent is an "io-unit" node.
3) Skip IRQ initialization when device lacks "reg" property.
Signed-off-by: David S. Miller <davem@davemloft.net>
Just name them "sbus%d" otherwise on sun4d we try to register
multiple entries named "sbi@0,0" which does not work.
Based upon a report from Raymond Burns.
Signed-off-by: David S. Miller <davem@davemloft.net>
On sparc32 the prom_{first,next}prop() interfaces work
a little differently. The buffer argument is ignored on
sparc32 and the firmware just returns a raw pointer to
the property name.
Signed-off-by: David S. Miller <davem@davemloft.net>
Table sizing code should look for "se" not "su" nodes.
The chip at the lower address should get the first index.
Signed-off-by: David S. Miller <davem@davemloft.net>
Register second SAB port before the first one, as serial A is wired to
it, and expected to appear as ttyS0.
Signed-off-by: Marc Zyngier <maz@misterjones.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sabre and Psycho PCI controllers can have partial interrupt-map
properties, meaning that on-board devices don't match up to any
entries. Instead, they are fully specified from the beginning and
we should pass them directly to the IRQ translator as-is.
Also, fill in the necessary translator slots for the "graphics"
and "expansion UPA" interrupts on Sabre, Psycho, and SYSIO SBUS.
Increase PROMREG_MAX to 24, as seen on SUNW,ffb devices.
Finally, prevent accidentally writing past the end of the of_device
struct resource[] and irqs[] arrays. Spit out a log message when
we ignore some entries because there are too many of them.
Signed-off-by: David S. Miller <davem@davemloft.net>
* master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6: (38 commits)
[SCSI] More buffer->request_buffer changes
[SCSI] mptfusion: bump version to 3.04.01
[SCSI] mptfusion: misc fix's
[SCSI] mptfusion: firmware download boot fix's
[SCSI] mptfusion: task abort fix's
[SCSI] mptfusion: sas nexus loss support
[SCSI] mptfusion: sas loginfo update
[SCSI] mptfusion: mptctl panic when loading
[SCSI] mptfusion: sas enclosures with smart drive
[SCSI] NCR_D700: misc fixes (section and argument ordering)
[SCSI] scsi_debug: must_check fixes
[SCSI] scsi_transport_sas: kill the use of channel
[SCSI] scsi_transport_sas: add expander backlink
[SCSI] hide EH backup data outside the scsi_cmnd
[SCSI] ibmvscsi: handle inactive SCSI target during probe
[SCSI] ibmvscsi: allocate lpevents for ibmvscsi on iseries
[SCSI] aic7[9x]xx: Remove last vestiges of reverse_scan
[SCSI] aha152x: stop poking at saved scsi_cmnd members
[SCSI] st.c: Improve sense output
[SCSI] lpfc 8.1.7: Change version number to 8.1.7
...
* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6:
[PATCH] spidernet: rework tx queue handling
[PATCH] spidernet: bug fix for init code
[PATCH] sky2: NAPI poll fix
[NET] ethtool: fix oops by testing correct struct member
e1000: bump version to 7.1.9-k4
e1000: fix panic on large frame receive when mtu=default
e1000: remove CRC bytes from measured packet length
e1000: Redo netpoll fix to address community concerns
Update ata_eh_about_to_do() and ata_eh_done() to improve EH action and
EHI flag handling.
* There are two types of EHI flags - one which expires on successful
EH and the other which expires on a successful reset. Make this
distinction clear.
* Unlike other EH actions, reset actions are represented by two EH
action masks and a EHI modifier. Implement correct about_to_do/done
semantics for resets. That is, prior to reset, related EH info is
sucked in from ehi and cleared, and after reset is complete, related
EH info in ehc is cleared.
These changes improve consistency and remove unnecessary EH actions
caused by stale EH action masks and EHI flags.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
* (ata_dev_absent() || ata_dev_ready()) test doesn't indicate
SUSPENDED state properly. Fix it.
* Link resuming resets shouldn't be skipped. Don't skip recovery on
EHI_RESUME_LINK. This doesn't matter for host ports as EHI_RESUME
always coincides with EHI_HOTPLUGGED which makes attached disabled
devices vacant. However, PMP reset causes non-hotplug link-resuming
resets which shouldn't be skipped.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Commit 0662c58b32 updated
ata_eh_autopsy() to OR determined action to ehc->i.action to preserve
action mask set directly into ehc->i.action by nested functions. This
broke action mask clearing on SENSE_VALID case causing revalidation
and EH complete message on successful ATAPI CC.
This patch removes two local variables - action and failed_dev - which
cache ehc->i.action and ehc->i.dev respectively, and make the function
directly modify ehc->i.* fields to remove aliasing issues.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Take return values of sysfs_create_group & friends into account.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
