Fernando Lopez-Lezcano reported frequent scheduling latencies and audio
xruns starting at the 2.6.18-rt kernel, and those problems persisted all
until current -rt kernels. The latencies were serious and unjustified by
system load, often in the milliseconds range.
After a patient and heroic multi-month effort of Fernando, where he
tested dozens of kernels, tried various configs, boot options,
test-patches of mine and provided latency traces of those incidents, the
following 'smoking gun' trace was captured by him:
_------=> CPU#
/ _-----=> irqs-off
| / _----=> need-resched
|| / _---=> hardirq/softirq
||| / _--=> preempt-depth
|||| /
||||| delay
cmd pid ||||| time | caller
\ / ||||| \ | /
IRQ_19-1479 1D..1 0us : __trace_start_sched_wakeup (try_to_wake_up)
IRQ_19-1479 1D..1 0us : __trace_start_sched_wakeup <<...>-5856> (37 0)
IRQ_19-1479 1D..1 0us : __trace_start_sched_wakeup (c01262ba 0 0)
IRQ_19-1479 1D..1 0us : resched_task (try_to_wake_up)
IRQ_19-1479 1D..1 0us : __spin_unlock_irqrestore (try_to_wake_up)
...
<idle>-0 1...1 11us!: default_idle (cpu_idle)
...
<idle>-0 0Dn.1 602us : smp_apic_timer_interrupt (c0103baf 1 0)
...
<...>-5856 0D..2 618us : __switch_to (__schedule)
<...>-5856 0D..2 618us : __schedule <<idle>-0> (20 162)
<...>-5856 0D..2 619us : __spin_unlock_irq (__schedule)
<...>-5856 0...1 619us : trace_stop_sched_switched (__schedule)
<...>-5856 0D..1 619us : trace_stop_sched_switched <<...>-5856> (37 0)
what is visible in this trace is that CPU#1 ran try_to_wake_up() for
PID:5856, it placed PID:5856 on CPU#0's runqueue and ran resched_task()
for CPU#0. But it decided to not send an IPI that no CPU - due to
TS_POLLING. But CPU#0 never woke up after its NEED_RESCHED bit was set,
and only rescheduled to PID:5856 upon the next lapic timer IRQ. The
result was a 600+ usecs latency and a missed wakeup!
the bug turned out to be an idle-wakeup bug introduced into the mainline
kernel this summer via an optimization in the x86_64 tree:
commit 495ab9c045
Author: Andi Kleen <ak@suse.de>
Date: Mon Jun 26 13:59:11 2006 +0200
[PATCH] i386/x86-64/ia64: Move polling flag into thread_info_status
During some profiling I noticed that default_idle causes a lot of
memory traffic. I think that is caused by the atomic operations
to clear/set the polling flag in thread_info. There is actually
no reason to make this atomic - only the idle thread does it
to itself, other CPUs only read it. So I moved it into ti->status.
the problem is this type of change:
if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
- clear_thread_flag(TIF_POLLING_NRFLAG);
+ current_thread_info()->status &= ~TS_POLLING;
smp_mb__after_clear_bit();
while (!need_resched()) {
local_irq_disable();
this changes clear_thread_flag() to an explicit clearing of TS_POLLING.
clear_thread_flag() is defined as:
clear_bit(flag, &ti->flags);
and clear_bit() is a LOCK-ed atomic instruction on all x86 platforms:
static inline void clear_bit(int nr, volatile unsigned long * addr)
{
__asm__ __volatile__( LOCK_PREFIX
"btrl %1,%0"
hence smp_mb__after_clear_bit() is defined as a simple compile barrier:
#define smp_mb__after_clear_bit() barrier()
but the explicit TS_POLLING clearing introduced by the patch:
+ current_thread_info()->status &= ~TS_POLLING;
is not an atomic op! So the clearing of the TS_POLLING bit is freely
reorderable with the reading of the NEED_RESCHED bit - and both now
reside in different memory addresses.
CPU idle wakeup very much depends on ordered memory ops, the clearing of
the TS_POLLING flag must always be done before we test need_resched()
and hit the idle instruction(s). [Symmetrically, the wakeup code needs
to set NEED_RESCHED before it tests the TS_POLLING flag, so memory
ordering is paramount.]
Fernando's dual-core Athlon64 system has a sufficiently advanced memory
ordering model so that it triggered this scenario very often.
( And it also turned out that the reason why these latencies never
triggered on my testsystems is that i routinely use idle=poll, which
was the only idle variant not affected by this bug. )
The fix is to change the smp_mb__after_clear_bit() to an smp_mb(), to
act as an absolute barrier between the TS_POLLING write and the
NEED_RESCHED read. This affects almost all idling methods (default,
ACPI, APM), on all 3 x86 architectures: i386, x86_64, ia64.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Tested-by: Fernando Lopez-Lezcano <nando@ccrma.Stanford.EDU>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
if CONFIG_CALGARY_IOMMU is built into the kernel via
CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT, or is enabled via the
iommu=calgary boot option, then the detect_calgary() function runs to
detect the presence of a Calgary IOMMU.
detect_calgary() first searches the BIOS EBDA area for a "rio_table_hdr"
BIOS table. It has this parsing algorithm for the EBDA:
while (offset) {
...
/* The next offset is stored in the 1st word. 0 means no more */
offset = *((unsigned short *)(ptr + offset));
}
got that? Lets repeat it slowly: we've got a BIOS-supplied data
structure, plus Linux kernel code that will only break out of an
infinite parsing loop once the BIOS gives a zero offset. Ok?
Translation: what an excellent opportunity for BIOS writers to lock up
the Linux boot process in an utterly hard to debug place! Indeed the
BIOS jumped on that opportunity on my box, which has the following EBDA
chaining layout:
384, 65282, 65535, 65535, 65535, 65535, 65535, 65535 ...
see the pattern? So my, definitely non-Calgary system happily locks up
in detect_calgary()!
the patch below fixes the boot hang by trusting the BIOS-supplied data
structure a bit less: the parser always has to make forward progress,
and if it doesnt, we break out of the loop and i get the expected kernel
message:
Calgary: Unable to locate Rio Grande Table in EBDA - bailing!
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
one of my boxes didnt boot the 2.6.20-rc1-rt0 kernel rpm, it hung during
early bootup. After an hour or two of happy debugging i narrowed it down
to the CALGARY_IOMMU_ENABLED_BY_DEFAULT option, which was freshly added
to 2.6.20 via the x86_64 tree and /enabled by default/.
commit bff6547bb6 claims:
[PATCH] Calgary: allow compiling Calgary in but not using it by default
This patch makes it possible to compile Calgary in but not use it by
default. In this mode, use 'iommu=calgary' to activate it.
but the change does not actually practice it:
config CALGARY_IOMMU_ENABLED_BY_DEFAULT
bool "Should Calgary be enabled by default?"
default y
depends on CALGARY_IOMMU
help
Should Calgary be enabled by default? if you choose 'y', Calgary
will be used (if it exists). If you choose 'n', Calgary will not be
used even if it exists. If you choose 'n' and would like to use
Calgary anyway, pass 'iommu=calgary' on the kernel command line.
If unsure, say Y.
it's both 'default y', and says "If unsure, say Y". Clearly not a typo.
disabling this option makes my box boot again. The patch below fixes the
Kconfig entry. Grumble.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
It has caused more problems than it ever really solved, and is
apparently not getting cleaned up and fixed. We can put it back when
it's stable and isn't likely to make warning or bug events worse.
In the meantime, enable frame pointers for more readable stack traces.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The new PDA code uses a dummy _proxy_pda variable to describe
memory references to the PDA. It is never referenced
in inline assembly, but exists as input/output arguments.
gcc 4.2 in some cases can CSE references to this which causes
unresolved symbols. Define it to zero to avoid this.
Signed-off-by: Andi Kleen <ak@suse.de>
2.6.19 stopped booting (or booted based on build/config) on our x86_64
systems due to a bug introduced in 2.6.19. check_nmi_watchdog schedules an
IPI on all cpus to busy wait on a flag, but fails to set the busywait
flag if NMI functionality is disabled. This causes the secondary cpus
to spin in an endless loop, causing the kernel bootup to hang.
Depending upon the build, the busywait flag got overwritten (stack variable)
and caused the kernel to bootup on certain builds. Following patch fixes
the bug by setting the busywait flag before returning from check_nmi_watchdog.
I guess using a stack variable is not good here as the calling function could
potentially return while the busy wait loop is still spinning on the flag.
AK: I redid the patch significantly to be cleaner
Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: Andi Kleen <ak@suse.de>
This facility provides three entry points:
ilog2() Log base 2 of unsigned long
ilog2_u32() Log base 2 of u32
ilog2_u64() Log base 2 of u64
These facilities can either be used inside functions on dynamic data:
int do_something(long q)
{
...;
y = ilog2(x)
...;
}
Or can be used to statically initialise global variables with constant values:
unsigned n = ilog2(27);
When performing static initialisation, the compiler will report "error:
initializer element is not constant" if asked to take a log of zero or of
something not reducible to a constant. They treat negative numbers as
unsigned.
When not dealing with a constant, they fall back to using fls() which permits
them to use arch-specific log calculation instructions - such as BSR on
x86/x86_64 or SCAN on FRV - if available.
[akpm@osdl.org: MMC fix]
Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: David Howells <dhowells@redhat.com>
Cc: Wojtek Kaniewski <wojtekka@toxygen.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Change all the uses of f_{dentry,vfsmnt} to f_path.{dentry,mnt} in the x86_64
arch code.
Signed-off-by: Josef "Jeff" Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This makes x86-64 use the generic BUG machinery.
The main advantage in using the generic BUG machinery for x86-64 is that
the inlined overhead of BUG is just the ud2a instruction; the file+line
information are no longer inlined into the instruction stream. This
reduces cache pollution.
Signed-off-by: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Andi Kleen <ak@muc.de>
Cc: Hugh Dickens <hugh@veritas.com>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The elf note saving code is currently duplicated over several
architectures. This cleanup patch simply adds code to a common file and
then replaces the arch-specific code with calls to the newly added code.
The only drawback with this approach is that s390 doesn't fully support
kexec-on-panic which for that arch leads to introduction of unused code.
Signed-off-by: Magnus Damm <magnus@valinux.co.jp>
Cc: Vivek Goyal <vgoyal@in.ibm.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
There was lots of #ifdef noise in the kernel due to hotcpu_notifier(fn,
prio) not correctly marking 'fn' as used in the !HOTPLUG_CPU case, and thus
generating compiler warnings of unused symbols, hence forcing people to add
#ifdefs.
the compiler can skip truly unused functions just fine:
text data bss dec hex filename
1624412 728710 3674856 6027978 5bfaca vmlinux.before
1624412 728710 3674856 6027978 5bfaca vmlinux.after
[akpm@osdl.org: topology.c fix]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
smp_call_function_single() can deadlock if the caller disabled local
interrupts (the target CPU could be spinning on call_lock). Check for that.
Why on earth do these functions use spin_lock_bh()??
Cc: "Randy.Dunlap" <rdunlap@xenotime.net>
Cc: Andi Kleen <ak@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When we are unregistering a kprobe-booster, we can't release its
instruction buffer immediately on the preemptive kernel, because some
processes might be preempted on the buffer. The freeze_processes() and
thaw_processes() functions can clean most of processes up from the buffer.
There are still some non-frozen threads who have the PF_NOFREEZE flag. If
those threads are sleeping (not preempted) at the known place outside the
buffer, we can ensure safety of freeing.
However, the processing of this check routine takes a long time. So, this
patch introduces the garbage collection mechanism of insn_slot. It also
introduces the "dirty" flag to free_insn_slot because of efficiency.
The "clean" instruction slots (dirty flag is cleared) are released
immediately. But the "dirty" slots which are used by boosted kprobes, are
marked as garbages. collect_garbage_slots() will be invoked to release
"dirty" slots if there are more than INSNS_PER_PAGE garbage slots or if
there are no unused slots.
Cc: "Keshavamurthy, Anil S" <anil.s.keshavamurthy@intel.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: "bibo,mao" <bibo.mao@intel.com>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Yumiko Sugita <yumiko.sugita.yf@hitachi.com>
Cc: Satoshi Oshima <soshima@redhat.com>
Cc: Hideo Aoki <haoki@redhat.com>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Define elf_addr_t in linux/elf.h. The size of the type is determined using
ELF_CLASS. This allows us to remove the defines that today are spread all
over .c and .h files.
Signed-off-by: Magnus Damm <magnus@valinux.co.jp>
Cc: Daniel Jacobowitz <drow@false.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Jakub Jelinek <jakub@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
After LOADER_TYPE && INITRD_START are true, the short if-condition
for INITRD_START can never be false.
Remove unused code from the else condition.
Signed-off-by: Henry Nestler <henry.ne@arcor.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The last thing we agreed on was to remove the macros entirely for 2.6.19,
on all architectures. Unfortunately, I think nobody actually _did_ that,
so they are still there.
[akpm@osdl.org: x86_64 fix]
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Greg Schafer <gschafer@zip.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
SLAB_KERNEL is an alias of GFP_KERNEL.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The .eh_frame section contents is never written to, so it can as well
benefit from CONFIG_DEBUG_RODATA.
Diff-ed against firstfloor tree.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
setup_IO_APIC_irqs could fail to get vector for some device when you have too
many devices, because at that time only boot cpu is online. So check vector
for irq in setup_ioapic_dest and call setup_IO_APIC_irq to make sure IO-APIC
irq-routing table is initialized.
Also seperate setup_IO_APIC_irq from setup_IO_APIC_irqs.
Signed-off-by: Yinghai Lu <yinghai.lu@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
- Remove "Disabling IOMMU" message because it confuses people
- Clarify that the GART IOMMU is refered to in other message
Signed-off-by: Andi Kleen <ak@suse.de>
Idle callbacks has some races when enter_idle() sets isidle and subsequent
interrupts that can happen on that CPU, before CPU goes to idle. Due to this,
an IDLE_END can get called before IDLE_START. To avoid these races, disable
interrupts before enter_idle and make sure that all idle routines do not
enable interrupts before entering idle.
Note that poll_idle() still has a this race as it has to enable interrupts
before going to idle. But, all other idle routines have the race fixed.
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
This was added as a workaround for the fallback unwinder not supporting
unaligned stack pointers properly. But now it was fixed to do that,
so it's not needed anymore
Cc: mingo@elte.hu
Signed-off-by: Andi Kleen <ak@suse.de>
Tighten the requirements on both input to and output from the Dwarf2
unwinder.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
We're already well protected against module unloads because module
unload uses stop_machine(). The only exception is NMIs, but other
users already risk lockless accesses here.
This avoids some hackery in lockdep and also a potential deadlock
This matches what i386 does.
Signed-off-by: Andi Kleen <ak@suse.de>
On the Core2 cpus, the rdtsc instruction is not serializing (as defined
in the architecture reference since rdtsc exists) and due to the deep
speculation of these cores, it's possible that you can observe time go
backwards between cores due to this speculation. Since the kernel
already deals with this with the SYNC_RDTSC flag, the solution is
simple, only assume that the instruction is serializing on family 15...
The price one pays for this is a slightly slower gettimeofday (by a
dozen or two cycles), but that increase is quite small to pay for a
really-going-forward tsc counter.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
There is no guarantee that two RDTSCs in a row are monotonic,
so don't assume it on single core AMD systems.
This will make gettimeofday slower again
Signed-off-by: Andi Kleen <ak@suse.de>
Make mce_remove_device() clean up the kobject in per_cpu(device_mce, cpu)
after it has been unregistered.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Andi Kleen <ak@suse.de>
interrupt array is referred for idt vectors instead of NR_IRQS, so change size
to NR_VECTORS - FIRST_EXTERNAL_VECTOR. Also change to static.
Signed-off-by: Yinghai Lu <yinghai@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
idt_table is in the .bss section, so clear_bss need to called at first
Signed-off-by: Yinghai Lu <yinghai.lu@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Add sysctl for kstack_depth_to_print. This lets users change
the amount of raw stack data printed in dump_stack() without
having to reboot.
Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Read/Write APIC_LVTPC and APIC_LVTTHMR only,
if get_maxlvt() returns certain values.
This is done like everywhere else in i386/kernel/apic.c,
so I guess its correct.
Suspends/Resumes to disk fine and eleminates an smp_error_interrupt()
here on a K8.
AK: ported to x86-64 too
Signed-off-by: Karsten Wiese <fzu@wemgehoertderstaat.de>
Signed-off-by: Andi Kleen <ak@suse.de>
Here is a small patch for x86-64 which adds a cpufeature flag and
detection code for Intel's Branch Trace Store (BTS) feature. This
feature can be found on Intel P4 and Core 2 processors among others.
It can also be used by perfmon.
changelog:
- add CPU_FEATURE_BTS
- add Branch Trace Store detection
signed-off-by: stephane eranian <eranian@hpl.hp.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Move the irqbalance quirks for E7320/E7520/E7525(Errata 23 in
http://download.intel.com/design/chipsets/specupdt/30304203.pdf) to early
quirks.
And add a PCI quirk for these platforms to check(which happens very late
during the boot) if the APIC routing is indeed set to default flat mode.
This fixes the breakage(in x86_64) of this quirk due to cpu hotplug which
selects physical mode instead of the logical flat(as needed for this errata
workaround).
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@suse.de>
Cc: "Li, Shaohua" <shaohua.li@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
The final line of /proc/<pid>/maps on x86_64 for native 64-bit
tasks shows an incorrect ending address and incorrect permissions. There
is only a single page mapped in this vsyscall region, and it is accessible
for both read and execute.
The patch below fixes this. (Since 32-bit-compat tasks have a real vma
with correct perms/range, no change is necessary for that scenario.)
Before the patch, a "cat /proc/self/maps | tail -1" shows this:
ffffffffff600000-ffffffffffe00000 ---p 00000000 [...]
After the patch, this is the output:
ffffffffff600000-ffffffffff601000 r-xp 00000000 [...]
Signed-off-by: Ernie Petrides <petrides@redhat.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Add an option to compile for Intel's Core 2
The Kconfig help is a mouthful due to the inventiveness of Intel's
product naming department.
Mainly for the 64bit cache line sizes because gcc doesn't support
optimizing for core2 yet. However it will and then the kernel
should be ready by passing the right option
Also fix the old MPSC help text to confirm better to reality.
Signed-off-by: Andi Kleen <ak@suse.de>
Port two patches from i386 to x86_64 delay.c to make sure all rounding is done
upward instead of downward.
There is no sign in commit messages that the mismatch was done on purpose, and
"delay() guarantees sleeping at least for the specified time" is still a valid
rule IMHO.
The original x86 patches are both from pre-GIT era, i.e.:
"[PATCH] round up in __udelay()" in commit
54c7e1f5cc6771ff644d7bc21a2b829308bd126f
"[PATCH] add 1 in __const_udelay()" in commit
42c77a9801b8877d8b90f65f75db758822a0bccc
(both commits are from converted BK repository to x86_64).
AK: fixed gcc warning
linux/arch/x86_64/lib/delay.c:43: warning: suggest parentheses around + or - inside shift
(did this actually work?)
Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andi Kleen <ak@suse.de>
