Fix atomicity of TIF update in flush_thread() for x86_64
Race :
parent process executing :
sys_ptrace()
(lock_kernel())
(ptrace_get_task_struct(pid))
arch_ptrace()
ptrace_detach()
ptrace_disable(child);
clear_singlestep(child);
clear_tsk_thread_flag(child, TIF_SINGLESTEP);
(which clears the TIF_SINGLESTEP flag atomically from a different
process)
(put_task_struct(child))
(unlock_kernel())
And at the same time, in the child process :
sys_execve()
do_execve()
search_binary_handler()
load_elf_binary()
flush_old_exec()
flush_thread()
doing a non-atomic thread flag update
Signed-off-by: Rebecca Schultz <rschultz@google.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Acked-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
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>
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>
Currently the idle loop has two nested loops -- one high level
in cpu_idle and in some low level idle functions another one.
Looping in the low level idle functions breaks the idle notifiers
because interrupts waking up sleep states need to execute
exit_idle() which is only in cpu_idle().
So don't do that, only loop in cpu_idle(). This only removes
code.
In some cases e.g. poll_idle the idle loop is a little longer
now because cpu_idle checks more things. I hope that isn't a problem
ACPI idle doesn't change behaviour because it never looped anyways.
Cc: len.brown@intel.com
Cc: eranian@hpl.hp.com
Signed-off-by: Andi Kleen <ak@suse.de>
The scheduler on Andreas Friedrich's hyperthreading system stopped
working properly: the scheduler would never move tasks to another CPU!
The lask known working kernel was 2.6.8.
After a couple of attempts to corner the bug, the following smoking gun
was found:
BIOS reported wrong ACPI idfor the processor
CPU#1: set_cpus_allowed(), swapper:1, 3 -> 2
[<c0103bbe>] show_trace_log_lvl+0x34/0x4a
[<c0103ceb>] show_trace+0x2c/0x2e
[<c01045f8>] dump_stack+0x2b/0x2d
[<c0116a77>] set_cpus_allowed+0x52/0xec
[<c0101d86>] cpu_idle_wait+0x2e/0x100
[<c0259c57>] acpi_processor_power_exit+0x45/0x58
[<c0259752>] acpi_processor_remove+0x46/0xea
[<c025c6fb>] acpi_start_single_object+0x47/0x54
[<c025cee5>] acpi_bus_register_driver+0xa4/0xd3
[<c04ab2d7>] acpi_processor_init+0x57/0x77
[<c01004d7>] init+0x146/0x2fd
[<c0103a87>] kernel_thread_helper+0x7/0x10
a quick look at cpu_idle_wait() shows how broken that code is
on i386: it changes the init task's affinity map but never
restores it ...
and because all userspace tasks get forked by init, they all
inherited that single-CPU affinity mask. x86_64 cloned this
bug too.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Andreas Friedrich <andreas.friedrich@fujitsu-siemens.com>
Cc: Wolfgang Erig <Wolfgang.Erig@fujitsu-siemens.com>
Cc: Andrew Morton <akpm@osdl.org>
Cc: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When another interrupt happens in exit_idle the exit idle notifier
could be called an incorrect number of times.
Add a test_and_clear_bit_pda and use it handle the bit
atomically against interrupts to avoid this.
Pointed out by Stephane Eranian
Signed-off-by: Andi Kleen <ak@suse.de>
Intel processors starting with the Core Duo support
support processor native C-state using the MWAIT instruction.
Refer: Intel Architecture Software Developer's Manual
http://www.intel.com/design/Pentium4/manuals/253668.htm
Platform firmware exports the support for Native C-state to OS using
ACPI _PDC and _CST methods.
Refer: Intel Processor Vendor-Specific ACPI: Interface Specification
http://www.intel.com/technology/iapc/acpi/downloads/302223.htm
With Processor Native C-state, we use 'MWAIT' instruction on the processor
to enter different C-states (C1, C2, C3). We won't use the special IO
ports to enter C-state and no SMM mode etc required to enter C-state.
Overall this will mean better C-state support.
One major advantage of using MWAIT for all C-states is, with this and
"treat interrupt as break event" feature of MWAIT, we can now get accurate
timing for the time spent in C1, C2, .. states.
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Len Brown <len.brown@intel.com>
This reverts an earlier patch that was found to cause FPU
state corruption. I think the corruption happens because
unlazy_fpu() can cause FPU exceptions and when it happens
after the current switch some processing would affect
the state in the wrong process.
Thanks to Douglas Crosher and Tom Hughes for testing.
Cc: jbeulich@novell.com
Signed-off-by: Andi Kleen <ak@suse.de>
In some places, particularly drivers and __init code, the init utsns is the
appropriate one to use. This patch replaces those with a the init_utsname
helper.
Changes: Removed several uses of init_utsname(). Hope I picked all the
right ones in net/ipv4/ipconfig.c. These are now changed to
utsname() (the per-process namespace utsname) in the previous
patch (2/7)
[akpm@osdl.org: CIFS fix]
Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Cc: Kirill Korotaev <dev@openvz.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: Andrey Savochkin <saw@sw.ru>
Cc: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Previously exit_idle would be called more often than enter_idle
Now instead of using complicated tests just keep track of it
using the per CPU variable as a flip flop. I moved the idle state into the
PDA to make the access more efficient.
Original bug report and an initial patch from Stephane Eranian,
but redone by AK.
Cc: Stephane Eranian <eranian@hpl.hp.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Fix
linux/arch/x86_64/kernel/process.c: In function __switch_to:
linux/arch/x86_64/kernel/process.c:626: warning: assignment makes integer from pointer without a cast
Signed-off-by: Andi Kleen <ak@suse.de>
This patch adds the per thread cookie field to the task struct and the PDA.
Also it makes sure that the PDA value gets the new cookie value at context
switch, and that a new task gets a new cookie at task creation time.
Signed-off-by: Arjan van Ven <arjan@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andi Kleen <ak@suse.de>
CC: Andi Kleen <ak@suse.de>
Right now the kernel on x86-64 has a 100% lazy fpu behavior: after *every*
context switch a trap is taken for the first FPU use to restore the FPU
context lazily. This is of course great for applications that have very
sporadic or no FPU use (since then you avoid doing the expensive
save/restore all the time). However for very frequent FPU users... you
take an extra trap every context switch.
The patch below adds a simple heuristic to this code: After 5 consecutive
context switches of FPU use, the lazy behavior is disabled and the context
gets restored every context switch. If the app indeed uses the FPU, the
trap is avoided. (the chance of the 6th time slice using FPU after the
previous 5 having done so are quite high obviously).
After 256 switches, this is reset and lazy behavior is returned (until
there are 5 consecutive ones again). The reason for this is to give apps
that do longer bursts of FPU use still the lazy behavior back after some
time.
[akpm@osdl.org: place new task_struct field next to jit_keyring to save space]
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Hello,
Following my discussion with Andi. Here is a patch that introduces
two new TIF flags to simplify the context switch code in __switch_to().
The idea is to minimize the number of cache lines accessed in the common
case, i.e., when neither the debug registers nor the I/O bitmap are used.
This patch covers the x86-64 modifications. A patch for i386 follows.
Changelog:
- add TIF_DEBUG to track when debug registers are active
- add TIF_IO_BITMAP to track when I/O bitmap is used
- modify __switch_to() to use the new TIF flags
<signed-off-by>: eranian@hpl.hp.com
Signed-off-by: Andi Kleen <ak@suse.de>
Only exports for assembler files are left in x8664_ksyms.c
Originally inspired by a patch from Al Viro
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
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.
Converted i386/x86-64/ia64 for now because that was the easiest
way to fix ACPI which also manipulates these flags in its idle
function.
Cc: Nick Piggin <npiggin@novell.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Len Brown <len.brown@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
These are the x86_64-specific pieces to enable reliable stack traces. The
only restriction with this is that it currently cannot unwind across the
interrupt->normal stack boundary, as that transition is lacking proper
annotation.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Previously it would only work in the first 32bit system call, not during
early process setup.
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Early development of x86-64 Linux was in CVS, but that hasn't been
the case for a long time now. Remove the obsolete $Id$s.
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
AMD K7/K8 CPUs only save/restore the FOP/FIP/FDP x87 registers in FXSAVE
when an exception is pending. This means the value leak through
context switches and allow processes to observe some x87 instruction
state of other processes.
This was actually documented by AMD, but nobody recognized it as
being different from Intel before.
The fix first adds an optimization: instead of unconditionally
calling FNCLEX after each FXSAVE test if ES is pending and skip
it when not needed. Then do a x87 load from a kernel variable to
clear FOP/FIP/FDP.
This means other processes always will only see a constant value
defined by the kernel in their FP state.
I took some pain to make sure to chose a variable that's already
in L1 during context switch to make the overhead of this low.
Also alternative() is used to patch away the new code on CPUs
who don't need it.
Patch for both i386/x86-64.
The problem was discovered originally by Jan Beulich. Richard
Brunner provided the basic code for the workarounds, with contribution
from Jan.
This is CVE-2006-1056
Cc: richard.brunner@amd.com
Cc: jbeulich@novell.com
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
In linux-2.6.16, we have noticed a problem where the gs base value
returned from an arch_prtcl(ARCH_GET_GS, ...) call will be incorrect if:
- the current/calling task has NOT set its own gs base yet to a
non-zero value,
- some other task that ran on the same processor previously set their
own gs base to a non-zero value.
In this situation, the ARCH_GET_GS code will read and return the
MSR_KERNEL_GS_BASE msr register.
However, since the __switch_to() code does NOT load/zero the
MSR_KERNEL_GS_BASE register when the task that is switched IN has a zero
next->gs value, the caller of arch_prctl(ARCH_GET_GS, ...) will get back
the value of some previous tasks's gs base value instead of 0.
Change the arch_prctl() ARCH_GET_GS code to only read and return
the MSR_KERNEL_GS_BASE msr register if the 'gs' register of the calling
task is non-zero.
Side note: Since in addition to using arch_prctl(ARCH_SET_GS, ...),
a task can also setup a gs base value by using modify_ldt() and write
an index value into 'gs' from user space, the patch below reads
'gs' instead of using thread.gs, since in the modify_ldt() case,
the thread.gs value will be 0, and incorrect value would be returned
(the task->thread.gs value).
When the user has not set its own gs base value and the 'gs'
register is zero, then the MSR_KERNEL_GS_BASE register will not be
read and a value of zero will be returned by reading and returning
'task->thread.gs'.
The first patch shown below is an attempt at implementing this
approach.
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When kretprobe probes the schedule() function, if the probed process exits
then schedule() will never return, so some kretprobe instances will never
be recycled.
In this patch the parent process will recycle retprobe instances of the
probed function and there will be no memory leak of kretprobe instances.
Signed-off-by: bibo mao <bibo.mao@intel.com>
Cc: Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Mathieu Desnoyers
Ingo Molnar
Venkatesh Pallipadi
Andi Kleen
Serge E. Hallyn
Arjan van de Ven
Stephane Eranian
Jan Beulich
John Blackwood
Alan Stern
bibo mao