X86 CPUs need to have some magic happening to enable the virtualization
extensions on them. This magic can result in unpleasant results for
users, like blocking other VMMs from working (vmx) or using invalid TLB
entries (svm).
Currently KVM activates virtualization when the respective kernel module
is loaded. This blocks us from autoloading KVM modules without breaking
other VMMs.
To circumvent this problem at least a bit, this patch introduces on
demand activation of virtualization. This means, that instead
virtualization is enabled on creation of the first virtual machine
and disabled on destruction of the last one.
So using this, KVM can be easily autoloaded, while keeping other
hypervisors usable.
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The only thing it protects now is interrupt injection into lapic and
this can work lockless. Even now with kvm->irq_lock in place access
to lapic is not entirely serialized since vcpu access doesn't take
kvm->irq_lock.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Maintain back mapping from irqchip/pin to gsi to speedup
interrupt acknowledgment notifications.
[avi: build fix on non-x86/ia64]
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This reverts commit b94b08081f.
genksyms currently cannot handle complicated types for exported
percpu variables. Drop this patch for now as it prevents a
module from being loaded on sn2 systems:
xpc: no symbol version for per_cpu____sn_cnodeid_to_nasid
xpc: Unknown symbol per_cpu____sn_cnodeid_to_nasid
Signed-off-by: Tony Luck <tony.luck@intel.com>
Use printk_once() in a couple of places.
Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
We had assumed that SAL firmware would return an error if it didn't
understand extended config space. Unfortunately, the SAL on the SGI 750
doesn't do that, it panics the machine. So, condition the extended PCI
config space accesses on SAL revision 3.2.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Tested-by: Brad Spengler <spender@grsecurity.net>
Signed-off-by: Tony Luck <tony.luck@intel.com>
struct sn_hwperf_ioctl_args member arg (u64) cannot be less than 0.
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Registers are not saved anywhere when INIT comes during fsys mode and
we cannot know what happened when we investigate vmcore captured by
kdump. This patch adds new function finish_pt_regs() so registers can
be saved in such a case.
Signed-off-by: Takao Indoh <indou.takao@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Bound the wait time for the ptcg_sem by using similar idea to the
ticket spin locks. In this case we have only one instance of a
spinaphore, so make it 8 bytes rather than try to squeeze it into
4-bytes to keep the code simpler (and shorter).
Signed-off-by: Tony Luck <tony.luck@intel.com>
Linus pointed out that other people have spent large amounts of time
and effort to optimize the layout of frequently used structures. Often
these have embedded locks, and the assumption is that a lock takes
4 bytes. Linus also pointed out how to work with the limited options
for atomic instructions on Itanium.
Signed-off-by: Tony Luck <tony.luck@intel.com>
* mark struct vm_area_struct::vm_ops as const
* mark vm_ops in AGP code
But leave TTM code alone, something is fishy there with global vm_ops
being used.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
ia64's sim_defconfig uses CONFIG_ACPI=n
which now #define's acpi_disabled in <linux/acpi.h>
So we shouldn't re-define it here in <asm/acpi.h>
Signed-off-by: Len Brown <len.brown@intel.com>
Back in January 2008 Nick Piggin implemented "ticket" spinlocks
for X86 (See commit 314cdbefd1).
IA64 implementation has a couple of differences because of the
available atomic operations ... e.g. we have no fetchadd2 instruction
that operates on a 16-bit quantity so we make ticket locks use
a 32-bit word for each of the current ticket and now-serving values.
Performance on uncontended locks is about 8% worse than the previous
implementation, but this seems a good trade for determinism in the
contended case. Performance impact on macro-level benchmarks is in
the noise.
Signed-off-by: Tony Luck <tony.luck@intel.com>
smp_call_function_many is the new version: it takes a pointer. Also,
use mm accessor macro while we're changing this.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
For /proc/kcore, each arch registers its memory range by kclist_add().
In usual,
- range of physical memory
- range of vmalloc area
- text, etc...
are registered but "range of physical memory" has some troubles. It
doesn't updated at memory hotplug and it tend to include unnecessary
memory holes. Now, /proc/iomem (kernel/resource.c) includes required
physical memory range information and it's properly updated at memory
hotplug. Then, it's good to avoid using its own code(duplicating
information) and to rebuild kclist for physical memory based on
/proc/iomem.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Some 64bit arch has special segment for mapping kernel text. It should be
entried to /proc/kcore in addtion to direct-linear-map, vmalloc area.
This patch unifies KCORE_TEXT entry scattered under x86 and ia64.
I'm not familiar with other archs (mips has its own even after this patch)
but range of [_stext ..._end) is a valid area of text and it's not in
direct-map area, defining CONFIG_ARCH_PROC_KCORE_TEXT is only a necessary
thing to do.
Note: I left mips as it is now.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For /proc/kcore, vmalloc areas are registered per arch. But, all of them
registers same range of [VMALLOC_START...VMALLOC_END) This patch unifies
them. By this. archs which have no kclist_add() hooks can see vmalloc
area correctly.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Presently, kclist_add() only eats start address and size as its arguments.
Considering to make kclist dynamically reconfigulable, it's necessary to
know which kclists are for System RAM and which are not.
This patch add kclist types as
KCORE_RAM
KCORE_VMALLOC
KCORE_TEXT
KCORE_OTHER
This "type" is used in a patch following this for detecting KCORE_RAM.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A number of architectures have identical asm/mman.h files so they can all
be merged by using the new generic file.
The remaining asm/mman.h files are substantially different from each
other.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a flag for mmap that will be used to request a huge page region that
will look like anonymous memory to user space. This is accomplished by
using a file on the internal vfsmount. MAP_HUGETLB is a modifier of
MAP_ANONYMOUS and so must be specified with it. The region will behave
the same as a MAP_ANONYMOUS region using small pages.
The patch also adds the MAP_STACK flag, which was previously defined only
on some architectures but not on others. Since MAP_STACK is meant to be a
hint only, architectures can define it without assigning a specific
meaning to it.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Eric B Munson <ebmunson@us.ibm.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: David Rientjes <rientjes@google.com>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 9617729941 ("Drop free_pages()")
modified nr_free_pages() to return 'unsigned long' instead of 'unsigned
int'. This made the casts to 'unsigned long' in most callers superfluous,
so remove them.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Kyle McMartin <kyle@mcmartin.ca>
Acked-by: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Mikael Starvik <starvik@axis.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Chris Zankel <zankel@tensilica.com>
Cc: Michal Simek <monstr@monstr.eu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Tim Abbott <tabbott@ksplice.com>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
ld-option is misnamed as it test options to gcc, not to ld.
Renamed it to reflect this.
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Replace the use of CROSS_COMPILE to select a customized
installkernel script with the possibility to set INSTALLKERNEL
to select a custom installkernel script when running make:
make INSTALLKERNEL=arm-installkernel install
With this patch we are now more consistent across
different architectures - they did not all support use
of CROSS_COMPILE.
The use of CROSS_COMPILE was a hack as this really belongs
to gcc/binutils and the installkernel script does not change
just because we change toolchain.
The use of CROSS_COMPILE caused troubles with an upcoming patch
that saves CROSS_COMPILE when a kernel is built - it would no
longer be installable.
[Thanks to Peter Z. for this hint]
This patch undos what Ian did in commit:
0f8e2d62fa
("use ${CROSS_COMPILE}installkernel in arch/*/boot/install.sh")
The patch has been lightly tested on x86 only - but all changes
looks obvious.
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Mike Frysinger <vapier@gentoo.org> [blackfin]
Acked-by: Russell King <linux@arm.linux.org.uk> [arm]
Acked-by: Paul Mundt <lethal@linux-sh.org> [sh]
Acked-by: "H. Peter Anvin" <hpa@zytor.com> [x86]
Cc: Ian Campbell <icampbell@arcom.com>
Cc: Tony Luck <tony.luck@intel.com> [ia64]
Cc: Fenghua Yu <fenghua.yu@intel.com> [ia64]
Cc: Hirokazu Takata <takata@linux-m32r.org> [m32r]
Cc: Geert Uytterhoeven <geert@linux-m68k.org> [m68k]
Cc: Kyle McMartin <kyle@mcmartin.ca> [parisc]
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> [powerpc]
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> [s390]
Cc: Thomas Gleixner <tglx@linutronix.de> [x86]
Cc: Ingo Molnar <mingo@redhat.com> [x86]
Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
The "cleanup console_print()" patch in commit
353f6dd2de introduced an "extern"
declaration into an assembly language file. Remove it.
Signed-off-by: Anirban Sinha <asinha@zeugmasystems.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Sysbench thinks SD_BALANCE_WAKE is too agressive and kbuild doesn't
really mind too much, SD_BALANCE_NEWIDLE picks up most of the
slack.
On a dual socket, quad core, dual thread nehalem system:
sysbench (--num_threads=16):
SD_BALANCE_WAKE-: 13982 tx/s
SD_BALANCE_WAKE+: 15688 tx/s
kbuild (-j16):
SD_BALANCE_WAKE-: 47.648295846 seconds time elapsed ( +- 0.312% )
SD_BALANCE_WAKE+: 47.608607360 seconds time elapsed ( +- 0.026% )
(same within noise)
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Aside from using fewer output sections and moving some data around,
the main side effect of this change is changing the alignment of some
sections. In particular:
* cachline-aligned and read_mostly data are now aligned to
SMP_CACHE_BYTES. (Previously, they were laid out consecutively after
a PAGE_SIZE alignment)
* .init.ramfs is now page-aligned, per the INIT_RAM_FS
macro. (Previously it had no explicit alignment).
Signed-off-by: Nelson Elhage <nelhage@ksplice.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
It seems that start_ap doesn't need to be in a special location in the
kernel, but it references some init code so it should be in .ref.text.
Since this is the only thing in the .text.head section, eliminate
.text.head from the linker script.
Signed-off-by: Tim Abbott <tabbott@ksplice.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
When
* there is almost out of ates
* one asks for more than one ate
* there are some available at the end of ate array
then the inner for loop will end without incrementing 'index'. This
means the outer loop will start at the same point finding it's available
and runs the inner loop again from the same index. This repeats forever.
Hence make sure we check we were at the end of ate array and return
an error in such case.
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Found-by: Jeff Mahoney <jeffm@novell.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
If we're looking to place a new task, we might as well find the
idlest position _now_, not 1 tick ago.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Make the idle balancer more agressive, to improve a
x264 encoding workload provided by Jason Garrett-Glaser:
NEXT_BUDDY NO_LB_BIAS
encoded 600 frames, 252.82 fps, 22096.60 kb/s
encoded 600 frames, 250.69 fps, 22096.60 kb/s
encoded 600 frames, 245.76 fps, 22096.60 kb/s
NO_NEXT_BUDDY LB_BIAS
encoded 600 frames, 344.44 fps, 22096.60 kb/s
encoded 600 frames, 346.66 fps, 22096.60 kb/s
encoded 600 frames, 352.59 fps, 22096.60 kb/s
NO_NEXT_BUDDY NO_LB_BIAS
encoded 600 frames, 425.75 fps, 22096.60 kb/s
encoded 600 frames, 425.45 fps, 22096.60 kb/s
encoded 600 frames, 422.49 fps, 22096.60 kb/s
Peter pointed out that this is better done via newidle_idx,
not via LB_BIAS, newidle balancing should look for where
there is load _now_, not where there was load 2 ticks ago.
Worst-case latencies are improved as well as no buddies
means less vruntime spread. (as per prior lkml discussions)
This change improves kbuild-peak parallelism as well.
Reported-by: Jason Garrett-Glaser <darkshikari@gmail.com>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1253011667.9128.16.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When merging select_task_rq_fair() and sched_balance_self() we lost
the use of wake_idx, restore that and set them to 0 to make wake
balancing more aggressive.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The problem with wake_idle() is that is doesn't respect things like
cpu_power, which means it doesn't deal well with SMT nor the recent
RT interaction.
To cure this, it needs to do what sched_balance_self() does, which
leads to the possibility of merging select_task_rq_fair() and
sched_balance_self().
Modify sched_balance_self() to:
- update_shares() when walking up the domain tree,
(it only called it for the top domain, but it should
have done this anyway), which allows us to remove
this ugly bit from try_to_wake_up().
- do wake_affine() on the smallest domain that contains
both this (the waking) and the prev (the wakee) cpu for
WAKE invocations.
Then use the top-down balance steps it had to replace wake_idle().
This leads to the dissapearance of SD_WAKE_BALANCE and
SD_WAKE_IDLE_FAR, with SD_WAKE_IDLE replaced with SD_BALANCE_WAKE.
SD_WAKE_AFFINE needs SD_BALANCE_WAKE to be effective.
Touch all topology bits to replace the old with new SD flags --
platforms might need re-tuning, enabling SD_BALANCE_WAKE
conditionally on a NUMA distance seems like a good additional
feature, magny-core and small nehalem systems would want this
enabled, systems with slow interconnects would not.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
console_print() is an old legacy interface mostly unused in the entire
kernel tree. It's best to clean up its existing use and let developers
use their own implementation of it as they feel fit.
Signed-off-by: Anirban Sinha <asinha@zeugmasystems.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Setting monarch_cpu = -1 to let slaves frozen might not work, because
there might be slaves being late, not entered the rendezvous yet.
Such slaves might be caught in while (monarch_cpu == -1) loop.
Use kdump_in_progress instead of monarch_cpus to break INIT rendezvous
and let all slaves enter DIE_INIT_SLAVE_LEAVE smoothly.
And monarch no longer need to manage rendezvous if once kdump_in_progress
is set, catch the monarch in DIE_INIT_MONARCH_ENTER then.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Haren Myneni <hbabu@us.ibm.com>
Cc: kexec@lists.infradead.org
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
kdump_on_init
CPUs should be frozen if possible, otherwise it might hinder kdump.
So if there are CPUs not respond to IPI, try INIT to stop them.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Haren Myneni <hbabu@us.ibm.com>
Cc: kexec@lists.infradead.org
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Summary:
Asserting INIT might block kdump if the system is already going to
start kdump via panic.
Description:
INIT can interrupt anywhere in panic path, so it can interrupt in
middle of kdump kicked by panic. Therefore there is a race if kdump
is kicked concurrently, via Panic and via INIT.
INIT could fail to invoke kdump if the system is already going to
start kdump via panic. It could not restart kdump from INIT handler
if some of cpus are already playing dead with INIT masked. It also
means that INIT could block kdump's progress if no monarch is entered
in the INIT rendezvous.
Panic+INIT is a rare, but possible situation since it can be assumed
that the kernel or an internal agent decides to panic the unstable
system while another external agent decides to send an INIT to the
system at same time.
How to reproduce:
Assert INIT just after panic, before all other cpus have frozen
Expected results:
continue kdump invoked by panic, or restart kdump from INIT
Actual results:
might be hang, crashdump not retrieved
Proposed Fix:
This patch masks INIT first in panic path to take the initiative on
kdump, and reuse atomic value kdump_in_progress to make sure there is
only one initiator of kdump. All INITs asserted later should be used
only for freezing all other cpus.
This mask will be removed soon by rfi in relocate_kernel.S, before jump
into kdump kernel, after all cpus are frozen and no-op INIT handler is
registered. So if INIT was in the interval while it is masked, it will
pend on the system and will received just after the rfi, and handled by
the no-op handler.
If there was a MCA event while psr.mc is 1, in theory the event will
pend on the system and will received just after the rfi same as above.
MCA handler is unregistered here at the time, so received MCA will not
reach to OS_MCA and will result in warmboot by SAL.
Note that codes in this masked interval are relatively simpler than
that in MCA/INIT handler which also executed with the mask. So it can
be said that probability of error in this interval is supposed not so
higher than that in MCA/INIT handler.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Haren Myneni <hbabu@us.ibm.com>
Cc: kexec@lists.infradead.org
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Summary:
Asserting INIT on cpu going to be offline will result in unexpected
behavior. It will be a real problem in kdump cases where INIT might
be asserted to unstable APs going to be offline by returning to SAL.
Description:
Since psr.mc is cleared when bits in psr are set to SAL_PSR_BITS_TO_SET
in ia64_jump_to_sal(), there is a small window (~few msecs) that the
cpu can receive INIT even if the cpu enter there via INIT handler.
In this window we do restore of registers for SAL, so INIT asserted
here will not work properly.
It is hard to remove this window by masking INIT (i.e. setting psr.mc)
because we have to unmask it later in OS, because we have to use branch
instruction (br.ret, not rfi) to return SAL, due to OS_BOOT_RENDEZ to
SAL return convention.
I suppose this window will not be a real problem on cpu offline if we
can educate people not to push INIT button during hotplug operation.
However, only exception is a race in kdump and INIT. Now kdump returns
APs to SAL before processing dump, but the kernel might receive INIT at
that point in time. Such INIT might be asserted by kdump itself if an
AP doesn't react IPI soon and kdump decided to use INIT to stop the AP.
Or it might be asserted by operator or an external agent to start dump
on the unstable system.
Such panic+INIT or INIT+INIT cases should be rare, but it will be happy
if we can retrieve crashdump even in such cases.
How to reproduce:
panic+INIT or INIT+INIT, with kdump configured
Expected results:
crashdump is retrieved anyway
Actual results:
panic, hang etc. (unexpected)
Proposed fix
To avoid the window on the way to SAL, this patch stops returning APs
to SAL in case of kdump. In other words, this patch makes APs spin
in OS instead of spinning in SAL.
(* Note: What impact would be there? If a cpu is spinning in SAL,
the cpu is in BOOT_RENDEZ loop, as same as offlined cpu.
In theory if an INIT is asserted there, cpus in the BOOT_RENDEZ loop
should not invoke OS_INIT on it. So in either way, no matter where
the cpu is spinning actually in, once cpu starts spin and act as
"frozen," INIT on the cpu have no effects.
From another point of view, all debug information on the cpu should
have stored to memory before the cpu start to be frozen. So no more
action on the cpu is required.)
I confirmed that the kdump sometime hangs by concurrent INITs (another
INIT after an INIT), and it doesn't hang after applying this patch.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Haren Myneni <hbabu@us.ibm.com>
Cc: kexec@lists.infradead.org
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Summary:
MCA on the beginning of kdump/kexec kernel will result in unexpected
behavior because MCA handler for previous kernel is invoked on the
kdump kernel.
Description:
Once a cpu is passed to new kernel, all resources in previous kernel
should not be used from the cpu. Even the resources for MCA handler
are no exception. So we cannot handle MCAs and its machine check
errors during kernel transition, until new handler for new kernel is
registered with new resources ready for handling the MCA.
How to reproduce:
Assert MCA while kdump kernel is booting, before new MCA handler for
kdump kernel is registered.
Expected(Desirable) results:
No recovery, cancel kdump and reboot the system.
Actual results:
MCA handler for previous kernel is invoked on the kdump kernel.
=> panic, hang etc. (unexpected)
Proposed fix:
To avoid entering MCA handler from early stage of new kernel,
unregister the entry point from SAL before leave from current
kernel. Then SAL will make all MCAs to warmboot safely, without
invoking OS_MCA.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Haren Myneni <hbabu@us.ibm.com>
Cc: kexec@lists.infradead.org
Acked-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Alexander Graf