[Xenomai] [PREVIEW] x86/ipipe queue for 3.5

[Xenomai] [PREVIEW] x86/ipipe queue for 3.5

[Jan Kiszka]

Hi,

at

    git://git.xenomai.org/ipipe-jki for-upstream/3.5

I've lined up our current x86 queue for I-pipe core-3.5. We are not yet
fully done with testing, I'd specifically like to give x86-32 a final
kick, so this is just a heads-up. The queue got a bit long:

Jan Kiszka (23):
      ipipe: Rework and simplify __ipipe_pin_vma
      ipipe: Fault in locked vmas after changing the protection flags
# critical fix for zero-page corruption after mprotect,
# actually affecting all mmu-archs

      x86/ipipe: Fix registration of IRQ work vector
      ipipe: Add root domain guard to preempt_schedule
      ipipe: Make context in add/sub_preempt_count conditional
      ipipe: Harden clocks used by ftrace and perf
      ipipe: Harden ftrace function and function-graph tracer
      ipipe: Harden trace_vbprintk for non-root use
      x86/ipipe: Prevent copy_from_user_nmi over non-root domains
      x86/ipipe: x86_64 provides context-agnostic current_thread_info
# ftrace and perf support, tested on x86-64 only,
# some bits may work on other archs too
# Note: I didn't follow the ipipe_safe_preempt_* path but rather disable
# preempt_count instrumentation selectively now. I think this is more
# appropriate given that the issue is only legacy related.

      x86/ipipe: Disallow transparent huge pages when I-pipe is active
# if you copy a distro .config, this can bite you... (we need to account for THP later)

      ipipe: Allow raw_spin_is_locked for I-pipe spin locks
      ipipe: Provide hardened probe_kernel_read/write for kgdb-over-I-pipe
      x86/ipipe: Implement arch-specific probe_kernel_read/write support
      x86/ipipe: Call kgdb_handle_exception over any domain in __ipipe_divert_exception
      x86/ipipe: Drop kgdb exception hook from from __ipipe_handle_exception
      ipipe: Harden generic part of kgdb for I-pipe use
      x86/ipipe: Harden kgdb for I-pipe use
      x86/ipipe: Fix kgdb hook invocation from __ipipe_divert_exception
# kgdb for x86 (tested on -64 only), may lay the ground for other archs too

      x86/ipipe: Add missing local_irq_enable on canceled vmentry
      ipipe: Fix IPIPE_HAVE_VM_NOTIFIER config option
      ipipe: Only compile __ipipe_notify_vm_preemption when needed
      ipipe: Relax context check for accessing vm_notifier in KVM callbacks
# get KVM working again on x86-64 (x86-32 is not recommended, even with vanilla)

 arch/x86/Kconfig                     |    1 +
 arch/x86/include/asm/hw_irq.h        |    2 +-
 arch/x86/kernel/ipipe.c              |   88 ++++++++++++++-------------------
 arch/x86/kernel/kgdb.c               |   24 +++++----
 arch/x86/kvm/x86.c                   |    1 +
 arch/x86/lib/usercopy.c              |    2 +-
 include/linux/ipipe.h                |   12 ++++-
 include/linux/ipipe_base.h           |    9 ++++
 include/linux/preempt.h              |    4 +-
 include/linux/spinlock.h             |    4 +-
 kernel/debug/debug_core.c            |   32 +++++++-----
 kernel/debug/gdbstub.c               |    6 +-
 kernel/ipipe/Kconfig                 |    6 ++
 kernel/ipipe/core.c                  |   39 +++++++++++++++
 kernel/sched/clock.c                 |   10 ++--
 kernel/sched/core.c                  |    5 +-
 kernel/trace/trace.c                 |    6 +-
 kernel/trace/trace_clock.c           |    4 +-
 kernel/trace/trace_functions.c       |    8 ++--
 kernel/trace/trace_functions_graph.c |    8 ++--
 mm/Kconfig                           |    1 +
 mm/memory.c                          |   79 ------------------------------
 mm/mlock.c                           |   18 +++++++
 mm/mprotect.c                        |   36 +++-----------
 24 files changed, 193 insertions(+), 212 deletions(-)

If there is time, please have a look, specifically at generic bits. Once
testing is done and there are no remarks, I'll send a pull request.

Thanks,
Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux

Re: [Xenomai] [PREVIEW] x86/ipipe queue for 3.5

[Jan Kiszka]

On 2013-01-11 19:13, Jan Kiszka wrote:
> Hi,
> 
> at
> 
>     git://git.xenomai.org/ipipe-jki for-upstream/3.5
> 
> I've lined up our current x86 queue for I-pipe core-3.5. We are not yet
> fully done with testing, I'd specifically like to give x86-32 a final
> kick, so this is just a heads-up.

Yep, and -32 is broken. Food for next week.

Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux

Re: [Xenomai] [PREVIEW] x86/ipipe queue for 3.5

[Jan Kiszka]

On 2013-01-11 19:38, Jan Kiszka wrote:
> On 2013-01-11 19:13, Jan Kiszka wrote:
>> Hi,
>>
>> at
>>
>>     git://git.xenomai.org/ipipe-jki for-upstream/3.5
>>
>> I've lined up our current x86 queue for I-pipe core-3.5. We are not yet
>> fully done with testing, I'd specifically like to give x86-32 a final
>> kick, so this is just a heads-up.
> 
> Yep, and -32 is broken. Food for next week.

Fortunately, it turned out to be an issue of my virtual test
environment, not of I-pipe.

Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux

Re: [Xenomai] [PREVIEW] x86/ipipe queue for 3.5

[Gilles Chanteperdrix]

On 01/11/2013 07:13 PM, Jan Kiszka wrote:

>       x86/ipipe: Disallow transparent huge pages when I-pipe is active
> # if you copy a distro .config, this can bite you... (we need to account for THP later)


What is wrong with transparent huge pages?

-- 
                                                                Gilles.

Re: [Xenomai] [PREVIEW] x86/ipipe queue for 3.5

[Jan Kiszka]

On 2013-01-12 03:39, Gilles Chanteperdrix wrote:
> On 01/11/2013 07:13 PM, Jan Kiszka wrote:
> 
>>       x86/ipipe: Disallow transparent huge pages when I-pipe is active
>> # if you copy a distro .config, this can bite you... (we need to account for THP later)
> 
> 
> What is wrong with transparent huge pages?

I didn't look at all details yet: we do not account for huge pages while
unCOW'ing e.g. Also, we once had THP enabled in some test setup and
quickly got crashes. Once understood, it's probably trivial to resolve.
But that needs to be done first. Huge page support is on our todo list.

Jan


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 261 bytes
Desc: OpenPGP digital signature
URL: <http://www.xenomai.org/pipermail/xenomai/attachments/20130112/3900fd16/attachment.pgp>