Peter Zijlstra [Fri, 27 Jun 2008 11:41:38 +0000 (13:41 +0200)]
sched: incremental effective_load()
Increase the accuracy of the effective_load values.
Not only consider the current increment (as per the attempted wakeup), but
also consider the delta between when we last adjusted the shares and the
current situation.
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:35 +0000 (13:41 +0200)]
sched: update shares on wakeup
We found that the affine wakeup code needs rather accurate load figures
to be effective. The trouble is that updating the load figures is fairly
expensive with group scheduling. Therefore ratelimit the updating.
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:34 +0000 (13:41 +0200)]
sched: fix shares boost logic
In case the domain is empty, pretend there is a single task on each cpu, so
that together with the boost logic we end up giving 1/n shares to each
cpu.
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:30 +0000 (13:41 +0200)]
sched: fix task_h_load()
Currently task_h_load() computes the load of a task and uses that to either
subtract it from the total, or add to it.
However, removing or adding a task need not have any effect on the total load
at all. Imagine adding a task to a group that is local to one cpu - in that
case the total load of that cpu is unaffected.
So properly compute addition/removal:
s_i = S * rw_i / \Sum_j rw_j
s'_i = S * (rw_i + wl) / (\Sum_j rw_j + wg)
then s'_i - s_i gives the change in load.
Where s_i is the shares for cpu i, S the group weight, rw_i the runqueue weight
for that cpu, wl the weight we add (subtract) and wg the weight contribution to
the runqueue.
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:29 +0000 (13:41 +0200)]
sched: fix load scaling in group balancing
doing the load balance will change cfs_rq->load.weight (that's the whole point)
but since that's part of the scale factor, we'll scale back with a different
amount.
Weight getting smaller would result in an inflated moved_load which causes
it to stop balancing too soon.
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:28 +0000 (13:41 +0200)]
sched: hierarchical load vs find_busiest_group
find_busiest_group() has some assumptions about task weight being in the
NICE_0_LOAD range. Hierarchical task groups break this assumption - fix this
by replacing it with the average task weight, which will adapt the situation.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:26 +0000 (13:41 +0200)]
sched: persistent average load per task
Remove the fall-back to SCHED_LOAD_SCALE by remembering the previous value of
cpu_avg_load_per_task() - this is useful because of the hierarchical group
model in which task weight can be much smaller.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:23 +0000 (13:41 +0200)]
sched: simplify the group load balancer
While thinking about the previous patch - I realized that using per domain
aggregate load values in load_balance_fair() is wrong. We should use the
load value for that CPU.
By not needing per domain hierarchical load values we don't need to store
per domain aggregate shares, which greatly simplifies all the math.
It basically falls apart in two separate computations:
- per domain update of the shares
- per CPU update of the hierarchical load
Also get rid of the move_group_shares() stuff - just re-compute the shares
again after a successful load balance.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:21 +0000 (13:41 +0200)]
sched: dont micro manage share losses
We used to try and contain the loss of 'shares' by playing arithmetic
games. Replace that by noticing that at the top sched_domain we'll
always have the full weight in shares to distribute.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
The idea was to balance groups until we've reached the global goal, however
Vatsa rightly pointed out that we might never reach that goal this way -
hence take out this logic.
[ the initial rationale for this 'feature' was to promote max concurrency
within a group - it does not however affect fairness ]
Reported-by: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:19 +0000 (13:41 +0200)]
sched: update aggregate when holding the RQs
It was observed that in __update_group_shares_cpu()
rq_weight > aggregate()->rq_weight
This is caused by forks/wakeups in between the initial aggregate pass and
locking of the RQs for load balance. To avoid this situation partially re-do
the aggregation once we have the RQs locked (which avoids new tasks from
appearing).
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:18 +0000 (13:41 +0200)]
sched: fix sched_domain aggregation
Keeping the aggregate on the first cpu of the sched domain has two problems:
- it could collide between different sched domains on different cpus
- it could slow things down because of the remote accesses
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:12 +0000 (13:41 +0200)]
sched: fix calc_delta_asym()
calc_delta_asym() is supposed to do the same as calc_delta_fair() except
linearly shrink the result for negative nice processes - this causes them
to have a smaller preemption threshold so that they are more easily preempted.
The problem is that for task groups se->load.weight is the per cpu share of
the actual task group weight; take that into account.
Also provide a debug switch to disable the asymmetry (which I still don't
like - but it does greatly benefit some workloads)
This would explain the interactivity issues reported against group scheduling.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra [Fri, 27 Jun 2008 11:41:10 +0000 (13:41 +0200)]
sched: clean up some unused variables
In file included from /mnt/build/linux-2.6/kernel/sched.c:1496:
/mnt/build/linux-2.6/kernel/sched_rt.c: In function '__enable_runtime':
/mnt/build/linux-2.6/kernel/sched_rt.c:339: warning: unused variable 'rd'
/mnt/build/linux-2.6/kernel/sched_rt.c: In function 'requeue_rt_entity':
/mnt/build/linux-2.6/kernel/sched_rt.c:692: warning: unused variable 'queue'
Signed-off-by: Peter Zijlstra <peterz@infradead.org> Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Linus Torvalds [Wed, 25 Jun 2008 01:12:33 +0000 (18:12 -0700)]
Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
[IA64] Eliminate NULL test after alloc_bootmem in iosapic_alloc_rte()
[IA64] Handle count==0 in sn2_ptc_proc_write()
[IA64] Fix boot failure on ia64/sn2
Linus Torvalds [Wed, 25 Jun 2008 01:09:06 +0000 (18:09 -0700)]
Merge branch 'kvm-updates-2.6.26' of git://git.kernel.org/pub/scm/linux/kernel/git/avi/kvm
* 'kvm-updates-2.6.26' of git://git.kernel.org/pub/scm/linux/kernel/git/avi/kvm:
KVM: Remove now unused structs from kvm_para.h
x86: KVM guest: Use the paravirt clocksource structs and functions
KVM: Make kvm host use the paravirt clocksource structs
x86: Make xen use the paravirt clocksource structs and functions
x86: Add structs and functions for paravirt clocksource
KVM: VMX: Fix host msr corruption with preemption enabled
KVM: ioapic: fix lost interrupt when changing a device's irq
KVM: MMU: Fix oops on guest userspace access to guest pagetable
KVM: MMU: large page update_pte issue with non-PAE 32-bit guests (resend)
KVM: MMU: Fix rmap_write_protect() hugepage iteration bug
KVM: close timer injection race window in __vcpu_run
KVM: Fix race between timer migration and vcpu migration
Jie Luo [Tue, 24 Jun 2008 17:38:31 +0000 (10:38 -0700)]
enable bus mastering on i915 at resume time
On 9xx chips, bus mastering needs to be enabled at resume time for much of the
chip to function. With this patch, vblank interrupts will work as expected
on resume, along with other chip functions. Fixes kernel bugzilla #10844.
Signed-off-by: Jie Luo <clotho67@gmail.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Gerd Hoffmann [Tue, 3 Jun 2008 14:17:32 +0000 (16:17 +0200)]
x86: KVM guest: Use the paravirt clocksource structs and functions
This patch updates the kvm host code to use the pvclock structs
and functions, thereby making it compatible with Xen.
The patch also fixes an initialization bug: on SMP systems the
per-cpu has two different locations early at boot and after CPU
bringup. kvmclock must take that in account when registering the
physical address within the host.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Gerd Hoffmann [Tue, 3 Jun 2008 14:17:29 +0000 (16:17 +0200)]
x86: Add structs and functions for paravirt clocksource
This patch adds structs for the paravirt clocksource ABI
used by both xen and kvm (pvclock-abi.h).
It also adds some helper functions to read system time and
wall clock time from a paravirtual clocksource (pvclock.[ch]).
They are based on the xen code. They are enabled using
CONFIG_PARAVIRT_CLOCK.
Subsequent patches of this series will put the code in use.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com> Acked-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch changes the computation for zero_metapath_length(), which it
renames to metapath_branch_start(). When you are extending the metadata
tree, The indirect blocks that point to the new data block must either
diverge from the existing tree either at the inode, or at the first
indirect block. They can diverge at the first indirect block because the
inode has room for 483 pointers while the indirect blocks have room for
509 pointers, so when the tree is grown, there is some free space in the
first indirect block. What metapath_branch_start() now computes is the
height where the first indirect block for the new data block is located.
It can either be 1 (if the indirect block diverges from the inode) or 2
(if it diverges from the first indirect block).
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Julia Lawall [Tue, 24 Jun 2008 08:22:05 +0000 (10:22 +0200)]
[IA64] Eliminate NULL test after alloc_bootmem in iosapic_alloc_rte()
As noted by Akinobu Mita alloc_bootmem and related functions never return
NULL and always return a zeroed region of memory. Thus a NULL test or
memset after calls to these functions is unnecessary.
Signed-off-by: Julia Lawall <julia@diku.dk> Signed-off-by: Tony Luck <tony.luck@intel.com>
Cliff Wickman [Tue, 24 Jun 2008 17:20:06 +0000 (10:20 -0700)]
[IA64] Handle count==0 in sn2_ptc_proc_write()
The fix applied in e0c6d97c65e0784aade7e97b9411f245a6c543e7
"security hole in sn2_ptc_proc_write" didn't take into account
the case where count==0 (which results in a buffer underrun
when adding the trailing '\0'). Thanks to Andi Kleen for
pointing this out.
Signed-off-by: Cliff Wickman <cpw@sgi.com> Signed-off-by: Tony Luck <tony.luck@intel.com>
Non-PAE operation has been deprecated in Xen for a while, and is
rarely tested or used. xen-unstable has now officially dropped
non-PAE support. Since Xen/pvops' non-PAE support has also been
broken for a while, we may as well completely drop it altogether.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Avi Kivity [Tue, 24 Jun 2008 08:48:49 +0000 (11:48 +0300)]
KVM: VMX: Fix host msr corruption with preemption enabled
Switching msrs can occur either synchronously as a result of calls to
the msr management functions (usually in response to the guest touching
virtualized msrs), or asynchronously when preempting a kvm thread that has
guest state loaded. If we're unlucky enough to have the two at the same
time, host msrs are corrupted and the machine goes kaput on the next syscall.
Most easily triggered by Windows Server 2008, as it does a lot of msr
switching during bootup.
Avi Kivity [Tue, 17 Jun 2008 22:36:36 +0000 (15:36 -0700)]
KVM: ioapic: fix lost interrupt when changing a device's irq
The ioapic acknowledge path translates interrupt vectors to irqs. It
currently uses a first match algorithm, stopping when it finds the first
redirection table entry containing the vector. That fails however if the
guest changes the irq to a different line, leaving the old redirection table
entry in place (though masked). Result is interrupts not making it to the
guest.
Fix by always scanning the entire redirection table.
Avi Kivity [Thu, 12 Jun 2008 13:54:41 +0000 (16:54 +0300)]
KVM: MMU: Fix oops on guest userspace access to guest pagetable
KVM has a heuristic to unshadow guest pagetables when userspace accesses
them, on the assumption that most guests do not allow userspace to access
pagetables directly. Unfortunately, in addition to unshadowing the pagetables,
it also oopses.
This never triggers on ordinary guests since sane OSes will clear the
pagetables before assigning them to userspace, which will trigger the flood
heuristic, unshadowing the pagetables before the first userspace access. One
particular guest, though (Xenner) will run the kernel in userspace, triggering
the oops. Since the heuristic is incorrect in this case, we can simply
remove it.
Marcelo Tosatti [Wed, 11 Jun 2008 23:32:40 +0000 (20:32 -0300)]
KVM: MMU: large page update_pte issue with non-PAE 32-bit guests (resend)
kvm_mmu_pte_write() does not handle 32-bit non-PAE large page backed
guests properly. It will instantiate two 2MB sptes pointing to the same
physical 2MB page when a guest large pte update is trapped.
Instead of duplicating code to handle this, disallow directory level
updates to happen through kvm_mmu_pte_write(), so the two 2MB sptes
emulating one guest 4MB pte can be correctly created by the page fault
handling path.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
rmap_next() does not work correctly after rmap_remove(), as it expects
the rmap chains not to change during iteration. Fix (for now) by restarting
iteration from the beginning.
Marcelo Tosatti [Fri, 6 Jun 2008 19:37:36 +0000 (16:37 -0300)]
KVM: close timer injection race window in __vcpu_run
If a timer fires after kvm_inject_pending_timer_irqs() but before
local_irq_disable() the code will enter guest mode and only inject such
timer interrupt the next time an unrelated event causes an exit.
It would be simpler if the timer->pending irq conversion could be done
with IRQ's disabled, so that the above problem cannot happen.
For now introduce a new vcpu requests bit to cancel guest entry.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Marcelo Tosatti [Fri, 6 Jun 2008 19:37:35 +0000 (16:37 -0300)]
KVM: Fix race between timer migration and vcpu migration
A guest vcpu instance can be scheduled to a different physical CPU
between the test for KVM_REQ_MIGRATE_TIMER and local_irq_disable().
If that happens, the timer will only be migrated to the current pCPU on
the next exit, meaning that guest LAPIC timer event can be delayed until
a host interrupt is triggered.
Fix it by cancelling guest entry if any vcpu request is pending. This
has the side effect of nicely consolidating vcpu->requests checks.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Linus Torvalds [Mon, 23 Jun 2008 23:25:11 +0000 (16:25 -0700)]
Merge branch 'hotfixes' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6
* 'hotfixes' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6:
NFS: nfs_updatepage(): don't mark page as dirty if an error occurred
NFS: Fix filehandle size comparisons in the mount code
NFS: Reduce the NFS mount code stack usage.
Linus Torvalds [Mon, 23 Jun 2008 19:48:50 +0000 (12:48 -0700)]
Merge branch 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: refactor wait_for_completion_timeout()
sched: fix wait_for_completion_timeout() spurious failure under heavy load
sched: rt: dont stop the period timer when there are tasks wanting to run
Linus Torvalds [Mon, 23 Jun 2008 19:48:17 +0000 (12:48 -0700)]
Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
xen: don't drop NX bit
xen: mask unwanted pte bits in __supported_pte_mask
xen: Use wmb instead of rmb in xen_evtchn_do_upcall().
x86: fix NULL pointer deref in __switch_to
Nick Piggin [Mon, 23 Jun 2008 12:30:30 +0000 (14:30 +0200)]
mm: fix race in COW logic
There is a race in the COW logic. It contains a shortcut to avoid the
COW and reuse the page if we have the sole reference on the page,
however it is possible to have two racing do_wp_page()ers with one
causing the other to mistakenly believe it is safe to take the shortcut
when it is not. This could lead to data corruption.
Process 1 and process2 each have a wp pte of the same anon page (ie.
one forked the other). The page's mapcount is 2. Then they both
attempt to write to it around the same time...
write private key into page
read from page
ptep_clear_flush()
set_pte_at(pte of new_page)
Fix this by moving the page_remove_rmap of the old page after the pte
clear and flush. Potentially the entire branch could be moved down
here, but in order to stay consistent, I won't (should probably move all
the *_mm_counter stuff with one patch).
Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
"This broke vmware 6.0.4.
Jun 22 14:53:03.845: vmx| NOT_IMPLEMENTED
/build/mts/release/bora-93057/bora/vmx/main/vmmonPosix.c:774"
and the reason seems to be that there's an old bug in how we handle do
FOLL_ANON on VM_SHARED areas in get_user_pages(), but since it only
triggered if the whole page table was missing, nobody had apparently hit
it before.
The recent changes to 'follow_page()' made the FOLL_ANON logic trigger
not just for whole missing page tables, but for individual pages as
well, and exposed this problem.
This fixes it by making the test for when FOLL_ANON is used more
careful, and also makes the code easier to read and understand by moving
the logic to a separate inline function.
Reported-and-tested-by: Jeff Chua <jeff.chua.linux@gmail.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Eli Cohen [Mon, 23 Jun 2008 16:29:58 +0000 (09:29 -0700)]
IB/mthca: Clear ICM pages before handing to FW
Current memfree FW has a bug which in some cases, assumes that ICM
pages passed to it are cleared. This patch uses __GFP_ZERO to
allocate all ICM pages passed to the FW. Once firmware with a fix is
released, we can make the workaround conditional on firmware version.
This fixes the bug reported by Arthur Kepner <akepner@sgi.com> here:
http://lists.openfabrics.org/pipermail/general/2008-May/050026.html
Cc: <stable@kernel.org> Signed-off-by: Eli Cohen <eli@mellanox.co.il>
[ Rewritten to be a one-liner using __GFP_ZERO instead of vmap()ing
ICM memory and memset()ing it to 0. - Roland ]
Thomas Gleixner [Mon, 23 Jun 2008 09:21:58 +0000 (11:21 +0200)]
futexes: fix fault handling in futex_lock_pi
This patch addresses a very sporadic pi-futex related failure in
highly threaded java apps on large SMP systems.
David Holmes reported that the pi_state consistency check in
lookup_pi_state triggered with his test application. This means that
the kernel internal pi_state and the user space futex variable are out
of sync. First we assumed that this is a user space data corruption,
but deeper investigation revieled that the problem happend because the
pi-futex code is not handling a fault in the futex_lock_pi path when
the user space variable needs to be fixed up.
The fault happens when a fork mapped the anon memory which contains
the futex readonly for COW or the page got swapped out exactly between
the unlock of the futex and the return of either the new futex owner
or the task which was the expected owner but failed to acquire the
kernel internal rtmutex. The current futex_lock_pi() code drops out
with an inconsistent in case it faults and returns -EFAULT to user
space. User space has no way to fixup that state.
When we wrote this code we thought that we could not drop the hash
bucket lock at this point to handle the fault.
After analysing the code again it turned out to be wrong because there
are only two tasks involved which might modify the pi_state and the
user space variable:
- the task which acquired the rtmutex
- the pending owner of the pi_state which did not get the rtmutex
Both tasks drop into the fixup_pi_state() function before returning to
user space. The first task which acquired the hash bucket lock faults
in the fixup of the user space variable, drops the spinlock and calls
futex_handle_fault() to fault in the page. Now the second task could
acquire the hash bucket lock and tries to fixup the user space
variable as well. It either faults as well or it succeeds because the
first task already faulted the page in.
One caveat is to avoid a double fixup. After returning from the fault
handling we reacquire the hash bucket lock and check whether the
pi_state owner has been modified already.
Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++-------------
1 file changed, 73 insertions(+), 20 deletions(-)
Linus Torvalds [Sun, 22 Jun 2008 19:23:15 +0000 (12:23 -0700)]
Fix performance regression on lmbench select benchmark
Christian Borntraeger reported that reinstating cond_resched() with
CONFIG_PREEMPT caused a performance regression on lmbench:
For example select file 500:
23 microseconds
32 microseconds
and that's really because we totally unnecessarily do the cond_resched()
in the innermost loop of select(), which is just silly.
This moves it out from the innermost loop (which only ever loops ove the
bits in a single "unsigned long" anyway), which makes the performance
regression go away.
Reported-and-tested-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The zonelist patches caused the loop that checks for available
objects in permitted zones to not terminate immediately. One object
per zone per allocation may be allocated and then abandoned.
Break the loop when we have successfully allocated one object.
Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
netns: Don't receive new packets in a dead network namespace.
sctp: Make sure N * sizeof(union sctp_addr) does not overflow.
pppoe: warning fix
ipv6: Drop packets for loopback address from outside of the box.
ipv6: Remove options header when setsockopt's optlen is 0
mac80211: detect driver tx bugs
Receiving packets while we are cleaning up a network namespace is a
racy proposition. It is possible when the packet arrives that we have
removed some but not all of the state we need to fully process it. We
have the choice of either playing wack-a-mole with the cleanup routines
or simply dropping packets when we don't have a network namespace to
handle them.
Since the check looks inexpensive in netif_receive_skb let's just
drop the incoming packets.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Fix warning:
drivers/net/pppoe.c: In function 'pppoe_recvmsg':
drivers/net/pppoe.c:945: warning: comparison of distinct pointer types lacks a cast
because skb->len is unsigned int and total_len is size_t
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ivan Kokshaysky [Fri, 20 Jun 2008 23:28:54 +0000 (03:28 +0400)]
alpha: resurrect Cypress IDE quirk
Which was removed in the hope that generic legacy IDE quirk in
drivers/pci/probe.c is sufficient for Cypress IDE.
It isn't, as this controller has non-standard BAR layout:
secondary channel registers are in the BAR0-1 of the second
PCI function - not in the BAR2-3 of the same function, as the
generic quirk routine assumes.
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ivan Kokshaysky [Fri, 20 Jun 2008 23:28:31 +0000 (03:28 +0400)]
alpha: fix compile failures with gcc-4.3 (bug #10438)
Vast majority of these build failures are gcc-4.3 warnings
about static functions and objects being referenced from
non-static (read: "extern inline") functions, in conjunction
with our -Werror.
We cannot just convert "extern inline" to "static inline",
as people keep suggesting all the time, because "extern inline"
logic is crucial for generic kernel build.
So
- just make sure that all callees of critical "extern inline"
functions are also "extern inline";
- use "static inline", wherever it's possible.
traps.c: work around gcc-4.3 being too smart about array
bounds-checking.
TODO: add "gnu_inline" attribute to all our "extern inline"
functions to ensure desired behaviour with future compilers.
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ivan Kokshaysky [Fri, 20 Jun 2008 23:26:21 +0000 (03:26 +0400)]
alpha: link failure fix
With built-in scsi disk driver, the final link fails with a following
error:
`.exit.text' referenced in section `.rodata' of drivers/built-in.o:
defined in discarded section `.exit.text' of drivers/built-in.o
This happens with -Os (CONFIG_CC_OPTIMIZE_FOR_SIZE=y) with all gcc-4
versions, and also with -O2 and gcc-4.3.
The problem is in sd.c:sd_major() being inlined into __exit function
exit_sd(), and the compiler generating a jump table in .rodata section
for the 'switch' statement in sd_major(). So we have references to
discarded section.
Fixed with a big hammer in the form of -fno-jump-tables.
Note that jump tables vs. discarded sections is a generic problem,
other architectures are just lucky not to suffer from it. But with
a slightly more complex switch/case statement it can be reproduced
on x86 as well. So maybe at some point we should consider
-fno-jump-tables as a generic compile option...
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ivan Kokshaysky [Fri, 20 Jun 2008 23:25:39 +0000 (03:25 +0400)]
alpha: fix module load failures on smp (bug #10926)
To calculate addresses of locally defined variables, GCC uses 32-bit
displacement from the GP. Which doesn't work for per cpu variables in
modules, as an offset to the kernel per cpu area is way above 4G.
The workaround is to force allocation of a GOT entry for per cpu variable
using ldq instruction with a 'literal' relocation.
I had to use custom asm/percpu.h, as a required argument magic doesn't
work with asm-generic/percpu.h macros.
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>