Patrick McManus [Fri, 21 Mar 2008 23:27:38 +0000 (16:27 -0700)]
[TCP]: TCP_DEFER_ACCEPT updates - defer timeout conflicts with max_thresh
timeout associated with SO_DEFER_ACCEPT wasn't being honored if it was
less than the timeout allowed by the maximum syn-recv queue size
algorithm. Fix by using the SO_DEFER_ACCEPT value if the ack has
arrived.
Signed-off-by: Patrick McManus <mcmanus@ducksong.com> Acked-by: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Pavel Emelyanov [Fri, 21 Mar 2008 22:52:00 +0000 (15:52 -0700)]
[NET]: NULL pointer dereference and other nasty things in /proc/net/(tcp|udp)[6]
Commits f40c81 ([NETNS][IPV4] tcp - make proc handle the network
namespaces) and a91275 ([NETNS][IPV6] udp - make proc handle the
network namespace) both introduced bad checks on sockets and tw
buckets to belong to proper net namespace.
I.e. when checking for socket to belong to given net and family the
do {
sk = sk_next(sk);
} while (sk && sk->sk_net != net && sk->sk_family != family);
constructions were used. This is wrong, since as soon as the
sk->sk_net fits the net the socket is immediately returned, even if it
belongs to other family.
As the result four /proc/net/(udp|tcp)[6] entries show wrong info.
The udp6 entry even oopses when dereferencing inet6_sk(sk) pointer:
dest = &np->daddr; /* will be NULL for AF_INET sockets */
...
seq_printf(...
dest->s6_addr32[0], dest->s6_addr32[1],
dest->s6_addr32[2], dest->s6_addr32[3],
...
Fix it by converting && to ||.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org> Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Make socket filters work for netlink unicast and notifications.
This is useful for applications like Zebra that get overrun with
messages that are then ignored.
Note: netlink messages are in host byte order, but packet filter
state machine operations are done as network byte order.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com> Signed-off-by: David S. Miller <davem@davemloft.net>
[NET]: Add per-connection option to set max TSO frame size
Update: My mailer ate one of Jarek's feedback mails... Fixed the
parameter in netif_set_gso_max_size() to be u32, not u16. Fixed the
whitespace issue due to a patch import botch. Changed the types from
u32 to unsigned int to be more consistent with other variables in the
area. Also brought the patch up to the latest net-2.6.26 tree.
Update: Made gso_max_size container 32 bits, not 16. Moved the
location of gso_max_size within netdev to be less hotpath. Made more
consistent names between the sock and netdev layers, and added a
define for the max GSO size.
Update: Respun for net-2.6.26 tree.
Update: changed max_gso_frame_size and sk_gso_max_size from signed to
unsigned - thanks Stephen!
This patch adds the ability for device drivers to control the size of
the TSO frames being sent to them, per TCP connection. By setting the
netdevice's gso_max_size value, the socket layer will set the GSO
frame size based on that value. This will propogate into the TCP
layer, and send TSO's of that size to the hardware.
This can be desirable to help tune the bursty nature of TSO on a
per-adapter basis, where one may have 1 GbE and 10 GbE devices
coexisting in a system, one running multiqueue and the other not, etc.
This can also be desirable for devices that cannot support full 64 KB
TSO's, but still want to benefit from some level of segmentation
offloading.
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jarek Poplawski [Fri, 21 Mar 2008 00:05:13 +0000 (17:05 -0700)]
[NET] ifb: set separate lockdep classes for queue locks
[ 10.536424] =======================================================
[ 10.536424] [ INFO: possible circular locking dependency detected ]
[ 10.536424] 2.6.25-rc3-devel #3
[ 10.536424] -------------------------------------------------------
[ 10.536424] swapper/0 is trying to acquire lock:
[ 10.536424] (&dev->queue_lock){-+..}, at: [<c0299b4a>]
dev_queue_xmit+0x175/0x2f3
[ 10.536424]
[ 10.536424] but task is already holding lock:
[ 10.536424] (&p->tcfc_lock){-+..}, at: [<f8a67154>] tcf_mirred+0x20/0x178
[act_mirred]
[ 10.536424]
[ 10.536424] which lock already depends on the new lock.
lockdep warns of locking order while using ifb with sch_ingress and
act_mirred: ingress_lock, tcfc_lock, queue_lock (usually queue_lock
is at the beginning). This patch is only to tell lockdep that ifb is
a different device (e.g. from eth) and has its own pair of queue
locks. (This warning is a false-positive in common scenario of using
ifb; yet there are possible situations, when this order could be
dangerous; lockdep should warn in such a case.) (With suggestions by
David S. Miller)
Reported-and-tested-by: Denys Fedoryshchenko <denys@visp.net.lb> Signed-off-by: Jarek Poplawski <jarkao2@gmail.com> Acked-by: Jamal Hadi Salim <hadi@cyberus.ca> Signed-off-by: David S. Miller <davem@davemloft.net>
Patrick McHardy [Thu, 20 Mar 2008 23:11:27 +0000 (16:11 -0700)]
[TCP]: Fix shrinking windows with window scaling
When selecting a new window, tcp_select_window() tries not to shrink
the offered window by using the maximum of the remaining offered window
size and the newly calculated window size. The newly calculated window
size is always a multiple of the window scaling factor, the remaining
window size however might not be since it depends on rcv_wup/rcv_nxt.
This means we're effectively shrinking the window when scaling it down.
The dump below shows the problem (scaling factor 2^7):
- Window size of 557 (71296) is advertised, up to 3111907257:
IP 172.2.2.3.33000 > 172.2.2.2.33000: . ack 3111835961 win 557 <...>
- New window size of 514 (65792) is advertised, up to 3111907217, 40 bytes
below the last end:
If the sender uses up the entire window before it is shrunk, this can have
chaotic effects on the connection. When sending ACKs, tcp_acceptable_seq()
will notice that the window has been shrunk since tcp_wnd_end() is before
tp->snd_nxt, which makes it choose tcp_wnd_end() as sequence number.
This will fail the receivers checks in tcp_sequence() however since it
is before it's tp->rcv_wup, making it respond with a dupack.
If both sides are in this condition, this leads to a constant flood of
ACKs until the connection times out.
Make sure the window is never shrunk by aligning the remaining window to
the window scaling factor.
Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
zap_completion_queue() retrieves skbs from completion_queue where they have
zero skb->users counter. Before dev_kfree_skb_any() it should be non-zero
yet, so it's increased now.
Reported-and-tested-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Jarek Poplawski <jarkao2@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Fabio Checconi [Thu, 20 Mar 2008 22:54:58 +0000 (15:54 -0700)]
bridge: use time_before() in br_fdb_cleanup()
In br_fdb_cleanup() next_timer and this_timer are in jiffies, so they
should be compared using the time_after() macro.
Signed-off-by: Fabio Checconi <fabio@gandalf.sssup.it> Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Pavel Emelyanov [Thu, 20 Mar 2008 22:39:41 +0000 (15:39 -0700)]
audit: netlink socket can be auto-bound to pid other than current->pid (v2)
From: Pavel Emelyanov <xemul@openvz.org>
This patch is based on the one from Thomas.
The kauditd_thread() calls the netlink_unicast() and passes
the audit_pid to it. The audit_pid, in turn, is received from
the user space and the tool (I've checked the audit v1.6.9)
uses getpid() to pass one in the kernel. Besides, this tool
doesn't bind the netlink socket to this id, but simply creates
it allowing the kernel to auto-bind one.
That's the preamble.
The problem is that netlink_autobind() _does_not_ guarantees
that the socket will be auto-bound to the current pid. Instead
it uses the current pid as a hint to start looking for a free
id. So, in case of conflict, the audit messages can be sent
to a wrong socket. This can happen (it's unlikely, but can be)
in case some task opens more than one netlink sockets and then
the audit one starts - in this case the audit's pid can be busy
and its socket will be bound to another id.
The proposal is to introduce an audit_nlk_pid in audit subsys,
that will point to the netlink socket to send packets to. It
will most often be equal to audit_pid. The socket id can be
got from the skb's netlink CB right in the audit_receive_msg.
The audit_nlk_pid reset to 0 is not required, since all the
decisions are taken based on audit_pid value only.
Later, if the audit tools will bind the socket themselves, the
kernel will have to provide a way to setup the audit_nlk_pid
as well.
A good side effect of this patch is that audit_pid can later
be converted to struct pid, as it is not longer safe to use
pid_t-s in the presence of pid namespaces. But audit code still
uses the tgid from task_struct in the audit_signal_info and in
the audit_filter_syscall.
Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: Pavel Emelyanov <xemul@openvz.org> Acked-by: Eric Paris <eparis@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Vlad Yasevich [Thu, 20 Mar 2008 22:17:14 +0000 (15:17 -0700)]
[SCTP]: Fix a race between module load and protosw access
There is a race is SCTP between the loading of the module
and the access by the socket layer to the protocol functions.
In particular, a list of addresss that SCTP maintains is
not initialized prior to the registration with the protosw.
Thus it is possible for a user application to gain access
to SCTP functions before everything has been initialized.
The problem shows up as odd crashes during connection
initializtion when we try to access the SCTP address list.
The solution is to refactor how we do registration and
initialize the lists prior to registering with the protosw.
Care must be taken since the address list initialization
depends on some other pieces of SCTP initialization. Also
the clean-up in case of failure now also needs to be refactored.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com> Acked-by: Sridhar Samudrala <sri@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
If a rule using ipt_recent is created with a hit count greater than
ip_pkt_list_tot, the rule will never match as it cannot keep track
of enough timestamps. This patch makes ipt_recent refuse to create such
rules.
With ip_pkt_list_tot's default value of 20, the following can be used
to reproduce the problem.
nc -u -l 0.0.0.0 1234 &
for i in `seq 1 100`; do echo $i | nc -w 1 -u 127.0.0.1 1234; done
This limits it to 20 packets:
iptables -A OUTPUT -p udp --dport 1234 -m recent --set --name test \
--rsource
iptables -A OUTPUT -p udp --dport 1234 -m recent --update --seconds \
60 --hitcount 20 --name test --rsource -j DROP
While this is unlimited:
iptables -A OUTPUT -p udp --dport 1234 -m recent --set --name test \
--rsource
iptables -A OUTPUT -p udp --dport 1234 -m recent --update --seconds \
60 --hitcount 21 --name test --rsource -j DROP
With the patch the second rule-set will throw an EINVAL.
Reported-by: Sean Kennedy <skennedy@vcn.com> Signed-off-by: Daniel Hokka Zakrisson <daniel@hozac.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
Linus Torvalds [Mon, 17 Mar 2008 16:52:24 +0000 (09:52 -0700)]
Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev
* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
ahci: Add Marvell 6121 SATA support
pata_ali: use atapi_cmd_type() to determine cmd type instead of transfer size
ahci: implement skip_host_reset parameter
ahci: request all PCI BARs
devres: implement pcim_iomap_regions_request_all()
libata-acpi: improve dock event handling
Tejun Heo [Tue, 11 Mar 2008 02:35:00 +0000 (11:35 +0900)]
pata_ali: use atapi_cmd_type() to determine cmd type instead of transfer size
pata_ali was using qc->nbytes to determine whether a command is
data transfer type or not. As now qc->nbytes can be extended by
padding and draining buffers, these tests are not useful anymore.
Use atapi_cmd_type() instead.
Signed-off-by: Tejun Heo <htejun@gmail.com> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: Rafael J. Wysocki <rjw@sisk.pl> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Tejun Heo [Mon, 10 Mar 2008 01:25:25 +0000 (10:25 +0900)]
ahci: implement skip_host_reset parameter
Under certain circumstances (SSP turned off by the BIOS) and for
debugging purposes, skipping global controller reset is helpful. Add
a kernel parameter for it.
Signed-off-by: Tejun Heo <htejun@gmail.com> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Tejun Heo [Tue, 11 Mar 2008 10:52:31 +0000 (19:52 +0900)]
ahci: request all PCI BARs
ahci is often implemented with accompanying SFF compatible interface
and legacy IDE driver may attach to the legacy IO ports when the
controller is already claimed by ahci and vice-versa. This patch
makes ahci use pcim_iomap_regions_request_all() so that all IO regions
are claimed on attach.
Signed-off-by: Tejun Heo <htejun@gmail.com> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Some drivers need to reserve all PCI BARs to prevent other drivers
misusing unoccupied BARs. pcim_iomap_regions_request_all() requests
all BARs and iomap specified BARs.
Signed-off-by: Tejun Heo <htejun@gmail.com> Cc: Greg Kroah-Hartman <gregkh@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: Jeff Garzik <jeff@garzik.org> Signed-off-by: Jeff Garzik <jeff@garzik.org>
ucc_geth: use correct thread number for 10/100Mbps link
Use thread number of 1 for 10/100Mbps link instead of 4.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se> Signed-off-by: Li Yang <leoli@freescale.com> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Ralf Baechle [Sat, 8 Mar 2008 16:58:33 +0000 (16:58 +0000)]
[IOC3] Fix section missmatch
LD drivers/net/built-in.o
WARNING: drivers/net/built-in.o(.text+0x3468): Section mismatch in reference fro
m the function ioc3_probe() to the function .devinit.text:ioc3_serial_probe()
The function ioc3_probe() references
the function __devinit ioc3_serial_probe().
This is often because ioc3_probe lacks a __devinit
annotation or the annotation of ioc3_serial_probe is wrong.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Ayaz Abdulla [Mon, 10 Mar 2008 19:58:21 +0000 (14:58 -0500)]
forcedeth: limit tx to 16
This is a critical patch which adds a workaround for a HW bug. The patch
will limit the number of outstanding tx packets to 16. Otherwise, the HW
could send out packets with bad checksums.
The driver will still setup the tx packets into the ring, however, will
only set the Valid bit on 16 packets at a time.
Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Divy Le Ray [Thu, 13 Mar 2008 07:13:30 +0000 (00:13 -0700)]
cxgb3: Fix transmit queue stop mechanism
The last change in the Tx queue stop mechanism opens a window
where the Tx queue might be stopped after pending credits
returned.
Tx credits are returned via a control message generated by the HW.
It returns tx credits on demand, triggered by a completion bit
set in selective transmit packet headers.
The current code can lead to the Tx queue stopped
with all pending credits returned, and the current frame
not triggering a credit return. The Tx queue will then never be
awaken.
The driver could alternatively request a completion for packets
that stop the queue. It's however safer at this point to go back
to the pre-existing behaviour.
Signed-off-by: Divy Le Ray <divy@chelsio.com> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Stefan Roese [Thu, 13 Mar 2008 15:59:43 +0000 (16:59 +0100)]
NEWEMAC: Add compatible "ibm,tah" to tah matching table
Add "ibm,tah" to the compatible matching table of the ibm_newemac
tah driver. The type "tah" is still preserved for compatibility reasons.
New dts files should use the compatible property though.
Signed-off-by: Stefan Roese <sr@denx.de> Signed-off-by: Jeff Garzik <jeff@garzik.org>
This patch should resolve a problem that's troubled support for
some RNDIS peripherals. It seems to have boiled down to using a
variable to establish transfer size limits before it was assigned,
which caused those devices to fallback to a default "jumbogram"
mode we don't support. Fix by assigning it earlier for RNDIS.
Signed-off-by: Jean-Christophe Dubois <jcd@tribudubois.net>
[ cleanups ] Signed-off-by: David Brownell <dbrownell@users.sourceforge.net> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Problem Description and Fix
---------------------------
When a pause packet(with destination as reserved Multicast address) is
received by the EMAC hardware to control the flow of frames being
transmitted by it, it is dropped by the hardware unless the reserved
Multicast address is hashed in to the GAHT[1-4] registers. This code fix
adds the default reserved multicast address to the GAHT[1-4] registers
in the EMAC(s) present on the chip. The flow control with Pause packets
will only work if the following register bits are programmed in EMAC:
EMACx_MR1[APP] = 1
EMACx_RMR[BAE] = 1
EMACx_RMR[MAE] = 1
Behavior that may be observed in a running system
-------------------------------------------------
A host transferring data from a PPC based system may send a Pause packet
to the PPC EMAC requesting it to slow down the flow of packets. If the
default reserved multicast MAC address is not programmed into the
GAHT[1-4] registers this Pause packet will be dropped by PPC EMAC and no
Flow Control will be done.
Signed-off-by: Pravin M. Bathija <pbathija@amcc.com> Signed-off-by: Stefan Roese <sr@denx.de> Signed-off-by: Jeff Garzik <jeff@garzik.org>
The problem can be triggered with a high amount of host->guest traffic.
I think its the following race:
poll says netif_rx_complete
poll calls enable_cb
enable_cb opens the interrupt mask
a new packet comes, an interrupt is triggered----\
enable_cb sees that there is more work |
enable_cb disables the interrupt |
. V
. interrupt is delivered
. skb_recv_done does atomic napi test, ok
some waiting disable_cb is called->check fails->bang!
.
poll would do napi check
poll would do disable_cb
The fix is to let enable_cb not disable the interrupt again, but expect the
caller to do the cleanup if it returns false. In that case, the interrupt is
only disabled, if the napi test_set_bit was successful.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (cleaned up doco)
Rusty Russell [Tue, 18 Mar 2008 03:58:15 +0000 (22:58 -0500)]
virtio: handle > 2 billion page balloon targets
If the host asks for a huge target towards_target() can overflow, and
we up oops as we try to release more pages than we have. The simple
fix is to use a 64-bit value.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Anthony Liguori [Sun, 2 Mar 2008 22:37:48 +0000 (16:37 -0600)]
virtio: Use spin_lock_irqsave/restore for virtio-pci
virtio-pci acquires its spin lock in an interrupt context so it's necessary
to use spin_lock_irqsave/restore variants. This patch fixes guest SMP when
using virtio devices in KVM.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Al Viro [Sun, 16 Mar 2008 22:22:04 +0000 (22:22 +0000)]
epic100 endianness annotations and fixes
* "powerpc or sparc" is not the same as "big-endian", fix the ifdef
* since we tell the card to byteswap the descriptors on big-endian,
we ought to leave them host-endian...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Helge Deller [Wed, 26 Dec 2007 17:07:01 +0000 (18:07 +0100)]
[PARISC] head.S: section mismatch fixes
- move boot_args[] into the init section
- move $global$ into the read_mostly section
- fix the following two section mismatches:
WARNING: vmlinux.o(.text+0x9c): Section mismatch: reference to .init.text:start_kernel (between '$pgt_fill_loop' and '$is_pa20')
WARNING: vmlinux.o(.text+0xa0): Section mismatch: reference to .init.text:start_kernel (between '$pgt_fill_loop' and '$is_pa20')
This is bogus on parisc, since page zero in kernel virtual space is the
gateway page for syscall entry, and should not be read from the kernel.
(That, and we really don't like the kernel faulting on its own address
space...)
Kyle McMartin [Sat, 1 Mar 2008 18:30:19 +0000 (10:30 -0800)]
[PARISC] clean up show_stack
When we show_regs, we obviously have a struct pt_regs of the calling
frame. Use these in show_stack so we don't have the entire bogus call trace
up to the show_stack call.
Adrian Bunk [Tue, 26 Feb 2008 19:55:17 +0000 (21:55 +0200)]
[PARISC] move defconfig to arch/parisc/configs/
This patch moves the default parisc defconfig to
arch/parisc/configs/generic_defconfig where it belongs and selects it as
the default defconfig through KBUILD_DEFCONFIG.
Signed-off-by: Adrian Bunk <adrian.bunk@movial.fi> Signed-off-by: Kyle McMartin <kyle@mcmartin.ca>
Kyle McMartin [Tue, 19 Feb 2008 07:34:34 +0000 (23:34 -0800)]
[PARISC] pdc_console: fix bizarre panic on boot
Commit 721fdf34167580ff98263c74cead8871d76936e6 introduced a subtle bug
by accidently removing the "static" from iodc_dbuf. This resulted in, what
appeared to be, a trap without *current set to a task. Probably the result of
a trap in real mode while calling firmware.
Also do other misc clean ups. Since the only input from firmware is non
blocking, share iodc_dbuf between input and output, and spinlock the
only callers.
Kyle McMartin [Tue, 19 Feb 2008 07:26:46 +0000 (23:26 -0800)]
[PARISC] dump_stack in show_regs
Originally, show_stack was used in BUG() output. However, a recent commit
changed it to print register state (no idea what that's supposed to help,
really...) and parisc was missing a backtrace because of it.
It did ugly things to the init sequence to populate the rootfs image
early, but that just ended up showing other problems with the whole
approach. The fact is, the VFS layer simply isn't initialized this
early, and the relevant ACPI code should either run much later, or this
shouldn't be done at all.
For 2.6.25, we'll just pick the latter option. We can revisit this
concept later if necessary.
Cc: Dave Hansen <haveblue@us.ibm.com> Cc: Tilman Schmidt <tilman@imap.cc> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Thomas Renninger <trenn@suse.de> Cc: Eric Piel <eric.piel@tremplin-utc.net> Cc: Len Brown <len.brown@intel.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Markus Gaugusch <dsdt@gaugusch.at> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ingo Molnar [Fri, 14 Mar 2008 21:17:08 +0000 (22:17 +0100)]
sched: simplify sched_slice()
Use the existing calc_delta_mine() calculation for sched_slice(). This
saves a divide and simplifies the code because we share it with the
other /cfs_rq->load users.
It also improves code size:
text data bss dec hex filename
42659 2740 144 45543 b1e7 sched.o.before
42093 2740 144 44977 afb1 sched.o.after
Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Peter Zijlstra [Fri, 14 Mar 2008 20:12:12 +0000 (21:12 +0100)]
sched: fix overload performance: buddy wakeups
Currently we schedule to the leftmost task in the runqueue. When the
runtimes are very short because of some server/client ping-pong,
especially in over-saturated workloads, this will cycle through all
tasks trashing the cache.
Reduce cache trashing by keeping dependent tasks together by running
newly woken tasks first. However, by not running the leftmost task first
we could starve tasks because the wakee can gain unlimited runtime.
Therefore we only run the wakee if its within a small
(wakeup_granularity) window of the leftmost task. This preserves
fairness, but does alternate server/client task groups.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 14 Mar 2008 19:55:51 +0000 (20:55 +0100)]
sched: min_vruntime fix
Current min_vruntime tracking is incorrect and will cause serious
problems when we don't run the leftmost task for some reason.
min_vruntime does two things; 1) it's used to determine a forward
direction when the u64 vruntime wraps, 2) it's used to track the
leftmost vruntime to position newly enqueued tasks from.
The current logic advances min_vruntime whenever the current task's
vruntime advance. Because the current task may pass the leftmost task
still waiting we're failing the second goal. This causes new tasks to be
placed too far ahead and thus penalizes their runtime.
Fix this by making min_vruntime the min_vruntime of the waiting tasks by
tracking it in enqueue/dequeue, and compare against current's vruntime
to obtain the absolute minimum when placing new tasks.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fix a hard to trigger crash seen in the -rt kernel that also affects
the vanilla scheduler.
There is a race condition between schedule() and some dequeue/enqueue
functions; rt_mutex_setprio(), __setscheduler() and sched_move_task().
When scheduling to idle, idle_balance() is called to pull tasks from
other busy processor. It might drop the rq lock. It means that those 3
functions encounter on_rq=0 and running=1. The current task should be
put when running.
The current process of CPU1(P1) is scheduling. Deactivated P1, and the
scheduler looks for another process on other CPU's runqueue because CPU1
will be idle. idle_balance(), load_balance_newidle() and
double_lock_balance() are called and double_lock_balance() could drop
the rq lock. On the other hand, CPU0 is trying to boost the priority of
P1. The result of boosting only P1's prio and sched_class are changed to
RT. The sched entities of P1 and P1's group are never put. It makes
cfs_rq invalid, because the cfs_rq has curr and no leaf, but
pick_next_task_fair() is called, then the kernel panics.
projects / linux-2.6-omap-h63xx.git / log