This patch allows a guest to use the VMXON and VMXOFF instructions, and
emulates them accordingly. Basically this amounts to checking some
prerequisites, and then remembering whether the guest has enabled or disabled
VMX operation.
Signed-off-by: Nadav Har'El <nyh@il.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch adds to kvm_intel a module option "nested". This option controls
whether the guest can use VMX instructions, i.e., whether we allow nested
virtualization. A similar, but separate, option already exists for the
SVM module.
This option currently defaults to 0, meaning that nested VMX must be
explicitly enabled by giving nested=1. When nested VMX matures, the default
should probably be changed to enable nested VMX by default - just like
nested SVM is currently enabled by default.
Signed-off-by: Nadav Har'El <nyh@il.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
During tracing the emulator, we noticed that init_emulate_ctxt()
sometimes took a bit longer time than we expected.
This patch is for mitigating the problem by some degree.
By looking into the function, we soon notice that it clears the whole
decode_cache whose size is about 2.5K bytes now. Furthermore, most of
the bytes are taken for the two read_cache arrays, which are used only
by a few instructions.
Considering the fact that we are not assuming the cache arrays have
been cleared when we store actual data, we do not need to clear the
arrays: 2K bytes elimination. In addition, we can avoid clearing the
fetch_cache and regs arrays.
This patch changes the initialization not to clear the arrays.
On our 64-bit host, init_emulate_ctxt() becomes 0.3 to 0.5us faster with
this patch applied.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Use a local pointer to the emulate_ctxt for simplicity. Then, arrange
the hard-to-read mode selection lines neatly.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
So far kvm_arch_vcpu_setup is responsible for freeing the vcpu struct if
it fails. Move this confusing resonsibility back into the hands of
kvm_vm_ioctl_create_vcpu. Only kvm_arch_vcpu_setup of x86 is affected,
all other archs cannot fail.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
In VMX, before we bring down a CPU we must VMCLEAR all VMCSs loaded on it
because (at least in theory) the processor might not have written all of its
content back to memory. Since a patch from June 26, 2008, this is done using
a per-cpu "vcpus_on_cpu" linked list of vcpus loaded on each CPU.
The problem is that with nested VMX, we no longer have the concept of a
vcpu being loaded on a cpu: A vcpu has multiple VMCSs (one for L1, a pool for
L2s), and each of those may be have been last loaded on a different cpu.
So instead of linking the vcpus, we link the VMCSs, using a new structure
loaded_vmcs. This structure contains the VMCS, and the information pertaining
to its loading on a specific cpu (namely, the cpu number, and whether it
was already launched on this cpu once). In nested we will also use the same
structure to hold L2 VMCSs, and vmx->loaded_vmcs is a pointer to the
currently active VMCS.
Signed-off-by: Nadav Har'El <nyh@il.ibm.com>
Acked-by: Acked-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Instead of blacklisting known-unsupported cpuid leaves, whitelist known-
supported leaves. This is more conservative and prevents us from reporting
features we don't support. Also whitelist a few more leaves while at it.
Signed-off-by: Avi Kivity <avi@redhat.com>
Acked-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Cleanup the same operation between kvm_mmu_page_unlink_children and
mmu_pte_write_zap_pte
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Parent pte rmap and page rmap are very similar, so use the same arithmetic
for them
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Abstract the operation of rmap to spte_list, then we can use it for the
reverse mapping of parent pte in the later patch
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Fix:
warning: ‘cs_sel’ may be used uninitialized in this function
warning: ‘ss_sel’ may be used uninitialized in this function
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Simply use __copy_to_user/__clear_user to write guest page since we have
already verified the user address when the memslot is set
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Simply return from kvm_mmu_pte_write path if no shadow page is
write-protected, then we can avoid to walk all shadow pages and hold
mmu-lock
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
vmcs_readl() and friends are really short, but gcc thinks they are long because of
the out-of-line exception handlers. Mark them always_inline to clear the
misunderstanding.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
We clean up a failed VMREAD by clearing the output register. Do
it in the exception handler instead of unconditionally. This is
worthwhile since there are more than a hundred call sites.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Dereference it in the actual users.
This not only cleans up the emulator but also makes it easy to convert
the old emulation functions to the new em_xxx() form later.
Note: Remove some inline keywords to let the compiler decide inlining.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Dereference it in the actual users: only do_insn_fetch_byte().
This is consistent with the way __linearize() dereferences it.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The two macros need special care to use:
Assume rc, ctxt, ops and done exist outside of them.
Can goto outside.
Considering the fact that these are used only in decode functions,
moving these right after do_insn_fetch() seems to be a right thing
to improve the readability.
We also rename do_fetch_insn_byte() to do_insn_fetch_byte() to be
consistent.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
* 'stable/bug.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/pci: Use the INT_SRC_OVR IRQ (instead of GSI) to preset the ACPI SCI IRQ.
xen/mmu: Fix for linker errors when CONFIG_SMP is not defined.
In the past we would use the GSI value to preset the ACPI SCI
IRQ which worked great as GSI == IRQ:
ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
While that is most often seen, there are some oddities:
ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
which means that GSI 20 (or pin 20) is to be overriden for IRQ 9.
Our code that presets the interrupt for ACPI SCI however would
use the GSI 20 instead of IRQ 9 ending up with:
xen: sci override: global_irq=20 trigger=0 polarity=1
xen: registering gsi 20 triggering 0 polarity 1
xen: --> pirq=20 -> irq=20
xen: acpi sci 20
.. snip..
calling acpi_init+0x0/0xbc @ 1
ACPI: SCI (IRQ9) allocation failed
ACPI Exception: AE_NOT_ACQUIRED, Unable to install System Control Interrupt handler (20110413/evevent-119)
ACPI: Unable to start the ACPI Interpreter
as the ACPI interpreter made a call to 'acpi_gsi_to_irq' which got nine.
It used that value to request an IRQ (request_irq) and since that was not
present it failed.
The fix is to recognize that for interrupts that are overriden (in our
case we only care about the ACPI SCI) we should use the IRQ number
to present the IRQ instead of the using GSI. End result is that we get:
xen: sci override: global_irq=20 trigger=0 polarity=1
xen: registering gsi 20 triggering 0 polarity 1
xen: --> pirq=20 -> irq=9 (gsi=9)
xen: acpi sci 9
which fixes the ACPI interpreter failing on startup.
CC: stable@kernel.org
Reported-by: Liwei <xieliwei@gmail.com>
Tested-by: Liwei <xieliwei@gmail.com>
[http://lists.xensource.com/archives/html/xen-devel/2011-06/msg01727.html]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Simple enough - we use an extern defined symbol which is not
defined when CONFIG_SMP is not defined. This fixes the linker
dying.
CC: stable@kernel.org
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
%rip-relative addressing is relative to the first byte of the next instruction,
so we need to add %rip only after we've fetched any immediate bytes.
Based on original patch by Li Xin <xin.li@intel.com>.
Signed-off-by: Avi Kivity <avi@redhat.com>
Acked-by: Li Xin <xin.li@intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Trying to build the Intel SCU Watchdog fails for me with gcc 4.6.0 -
$ gcc --version | head -n 1
gcc (GCC) 4.6.0 20110513 (prerelease)
like this :
CC drivers/watchdog/intel_scu_watchdog.o
In file included from drivers/watchdog/intel_scu_watchdog.c:49:0:
/home/jj/src/linux-2.6/arch/x86/include/asm/apb_timer.h: In function ‘apbt_time_init’:
/home/jj/src/linux-2.6/arch/x86/include/asm/apb_timer.h:65:42: warning: ‘return’ with a value, in function returning void [enabled by default]
drivers/watchdog/intel_scu_watchdog.c: In function ‘intel_scu_watchdog_init’:
drivers/watchdog/intel_scu_watchdog.c:468:2: error: implicit declaration of function ‘sfi_get_mtmr’ [-Werror=implicit-function-declaration]
drivers/watchdog/intel_scu_watchdog.c:468:32: warning: assignment makes pointer from integer without a cast [enabled by default]
cc1: some warnings being treated as errors
make[1]: *** [drivers/watchdog/intel_scu_watchdog.o] Error 1
make: *** [drivers/watchdog/intel_scu_watchdog.o] Error 2
Additionally, linux/types.h is needlessly being included twice in
drivers/watchdog/intel_scu_watchdog.c
Signed-off-by: Jesper Juhl <jj@chaosbits.net>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
commit 21a3c96 uses node_start/end_pfn(nid) for detection start/end
of nodes. But, it's not defined in linux/mmzone.h but defined in
/arch/???/include/mmzone.h which is included only under
CONFIG_NEED_MULTIPLE_NODES=y.
Then, we see
mm/page_cgroup.c: In function 'page_cgroup_init':
mm/page_cgroup.c:308: error: implicit declaration of function 'node_start_pfn'
mm/page_cgroup.c:309: error: implicit declaration of function 'node_end_pfn'
So, fixiing page_cgroup.c is an idea...
But node_start_pfn()/node_end_pfn() is a very generic macro and
should be implemented in the same manner for all archs.
(m32r has different implementation...)
This patch removes definitions of node_start/end_pfn() in each archs
and defines a unified one in linux/mmzone.h. It's not under
CONFIG_NEED_MULTIPLE_NODES, now.
A result of macro expansion is here (mm/page_cgroup.c)
for !NUMA
start_pfn = ((&contig_page_data)->node_start_pfn);
end_pfn = ({ pg_data_t *__pgdat = (&contig_page_data); __pgdat->node_start_pfn + __pgdat->node_spanned_pages;});
for NUMA (x86-64)
start_pfn = ((node_data[nid])->node_start_pfn);
end_pfn = ({ pg_data_t *__pgdat = (node_data[nid]); __pgdat->node_start_pfn + __pgdat->node_spanned_pages;});
Changelog:
- fixed to avoid using "nid" twice in node_end_pfn() macro.
Reported-and-acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Reported-and-tested-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes/pci-2.6:
x86/PCI/ACPI: fix type mismatch
PCI: fix new kernel-doc warning
PCI: Fix warning in drivers/pci/probe.c on sparc64
* 'stable/bug.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/setup: Fix for incorrect xen_extra_mem_start.
xen: When calling power_off, don't call the halt function.
xen: Fix compile warning when CONFIG_SMP is not defined.
xen: support CONFIG_MAXSMP
xen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"
* 'kvm-updates/3.0' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: Fix register corruption in pvclock_scale_delta
KVM: MMU: fix opposite condition in mapping_level_dirty_bitmap
KVM: VMX: do not overwrite uptodate vcpu->arch.cr3 on KVM_SET_SREGS
KVM: MMU: Fix build warnings in walk_addr_generic()
The 128-bit multiply in pvclock.h was missing an output constraint for
EDX which caused a register corruption to appear. Thanks to Ulrich for
diagnosing the EDX corruption and Avi for providing this fix.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The condition is opposite, it always maps huge page for the dirty tracked page
Reported-by: Steve <stefan.bosak@gmail.com>
Signed-off-by: Steve <stefan.bosak@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Only decache guest CR3 value if vcpu->arch.cr3 is stale.
Fixes loadvm with live guest.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Tested-by: Markus Schade <markus.schade@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
On 3.0-rc1 I get
In file included from arch/x86/kvm/mmu.c:2856:
arch/x86/kvm/paging_tmpl.h: In function ‘paging32_walk_addr_generic’:
arch/x86/kvm/paging_tmpl.h:124: warning: ‘ptep_user’ may be used uninitialized in this function
In file included from arch/x86/kvm/mmu.c:2852:
arch/x86/kvm/paging_tmpl.h: In function ‘paging64_walk_addr_generic’:
arch/x86/kvm/paging_tmpl.h:124: warning: ‘ptep_user’ may be used uninitialized in this function
caused by 6e2ca7d1802bf8ed9908435e34daa116662e7790. According to Takuya
Yoshikawa, ptep_user won't be used uninitialized so shut up gcc.
Cc: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Link: http://lkml.kernel.org/r/20110530094604.GC21833@liondog.tnic
Signed-off-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
Commit 916f676f8dc started reserving boot service code since some systems
require you to keep that code around until SetVirtualAddressMap is called.
However, in some cases those areas will overlap with reserved regions.
The proper medium-term fix is to fix the bootloader to prevent the
conflicts from occurring by moving the kernel to a better position,
but the kernel should check for this possibility, and only reserve regions
which can be reserved.
Signed-off-by: Maarten Lankhorst <m.b.lankhorst@gmail.com>
Link: http://lkml.kernel.org/r/4DF7A005.1050407@gmail.com
Acked-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The earlier attempts (24bdb0b62cc82120924762ae6bc85afc8c3f2b26)
at fixing this problem caused other problems to surface (PV guests
with no PCI passthrough would have SWIOTLB turned on - which meant
64MB of precious contingous DMA32 memory being eaten up per guest).
The problem was: "on xen we add an extra memory region at the end of
the e820, and on this particular machine this extra memory region
would start below 4g and cross over the 4g boundary:
[0xfee01000-0x192655000)
Unfortunately e820_end_of_low_ram_pfn does not expect an
e820 layout like that so it returns 4g, therefore initial_memory_mapping
will map [0 - 0x100000000), that is a memory range that includes some
reserved memory regions."
The memory range was the IOAPIC regions, and with the 1-1 mapping
turned on, it would map them as RAM, not as MMIO regions. This caused
the hypervisor to complain. Fortunately this is experienced only under
the initial domain so we guard for it.
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
.. As it won't actually power off the machine.
Reported-by: Sven Köhler <sven.koehler@gmail.com>
Tested-by: Sven Köhler <sven.koehler@gmail.com>
Signed-off-by: Tom Goetz <tom.goetz@virtualcomputer.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The MAXSMP config option requires CPUMASK_OFFSTACK, which in turn
requires we init the memory for the maps while we bring up the cpus.
MAXSMP also increases NR_CPUS to 4096. This increase in size exposed an
issue in the argument construction for multicalls from
xen_flush_tlb_others. The args should only need space for the actual
number of cpus.
Also in 2.6.39 it exposes a bootup problem.
BUG: unable to handle kernel NULL pointer dereference at (null)
IP: [<ffffffff8157a1d3>] set_cpu_sibling_map+0x123/0x30d
...
Call Trace:
[<ffffffff81039a3f>] ? xen_restore_fl_direct_reloc+0x4/0x4
[<ffffffff819dc4db>] xen_smp_prepare_cpus+0x36/0x135
..
CC: stable@kernel.org
Signed-off-by: Andrew Jones <drjones@redhat.com>
[v2: Updated to compile on 3.0]
[v3: Updated to compile when CONFIG_SMP is not defined]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
[ Also from Ben Hutchings <ben@decadent.org.uk> and Vitaliy Ivanov
<vitalivanov@gmail.com> ]
Commit 06ae40ce073d ("x86 idle: EXPORT_SYMBOL(default_idle, pm_idle)
only when APM demands it") removed the export for pm_idle/default_idle
unless the apm module was modularised and CONFIG_APM_CPU_IDLE was set.
But the apm module uses pm_idle/default_idle unconditionally,
CONFIG_APM_CPU_IDLE only affects the bios idle threshold. Adjust the
export accordingly.
[ Used #ifdef instead of #if defined() as it's shorter, and what both
Ben and Vitaliy used.. Andy, you're out-voted ;) - Linus ]
Reported-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Len Brown <len.brown@intel.com>
Signed-off-by: Andy Whitcroft <apw@canonical.com>
Signed-off-by: Vitaliy Ivanov <vitalivanov@gmail.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
drm: Compare only lower 32 bits of framebuffer map offsets
drm/i915: Don't leak in i915_gem_shmem_pread_slow()
drm/radeon/kms: do bounds checking for 3D_LOAD_VBPNTR and bump array limit
drm/radeon/kms: fix mac g5 quirk
x86/uv/x2apic: update for change in pci bridge handling.
alpha, drm: Remove obsolete Alpha support in MGA DRM code
alpha/drm: Cleanup Alpha support in DRM generic code
savage: remove unnecessary if statement
drm/radeon: fix GUI idle IH debug statements
drm/radeon/kms: check modes against max pixel clock
drm: fix fbs in DRM_IOCTL_MODE_GETRESOURCES ioctl
When I added 3448a19da479b6bd1e28e2a2be9fa16c6a6feb39
I forgot about the special uv handling code for this, so this
patch fixes it up.
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Ingo Molnar
Signed-off-by: Dave Airlie <airlied@redhat.com>
Unconditionally changing the address limit to USER_DS and not restoring
it to its old value in the error path is wrong because it prevents us
using kernel memory on repeated calls to this function. This, in fact,
breaks the fallback of hard coded paths to the init program from being
ever successful if the first candidate fails to load.
With this patch applied switching to USER_DS is delayed until the point
of no return is reached which makes it possible to have a multi-arch
rootfs with one arch specific init binary for each of the (hard coded)
probed paths.
Since the address limit is already set to USER_DS when start_thread()
will be invoked, this redundancy can be safely removed.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch fixes the following build failure:
drivers/built-in.o: In function `early_init_dt_check_for_initrd':
/home/florian/dev/kernel/x86/linux-2.6-x86/drivers/of/fdt.c:571:
undefined reference to `early_init_dt_setup_initrd_arch'
make: *** [.tmp_vmlinux1] Error 1
which happens as soon as we enable initrd support on a x86 devicetree
platform such as Intel CE4100.
Signed-off-by: Florian Fainelli <ffainelli@freebox.fr>
Acked-by: Grant Likely <grant.likely@secretlab.ca>
Cc: Maxime Bizon <mbizon@freebox.fr>
Acked-by: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: stable@kernel.org # 2.6.39
Link: http://lkml.kernel.org/r/201106061015.50039.ffainelli@freebox.fr
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
We only need to set max_pfn_mapped to the last pfn mapped on x86_64 to
make sure that cleanup_highmap doesn't remove important mappings at
_end.
We don't need to do this on x86_32 because cleanup_highmap is not called
on x86_32. Besides lowering max_pfn_mapped on x86_32 has the unwanted
side effect of limiting the amount of memory available for the 1:1
kernel pagetable allocation.
This patch reverts the x86_32 part of the original patch.
CC: stable@kernel.org
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
* 'stable/bug.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen: off by one errors in multicalls.c
xen: use the trigger info we already have to choose the irq handler
After a newly plugged CPU sets the cpu_online bit it enables
interrupts and goes idle. The cpu which brought up the new cpu waits
for the cpu_online bit and when it observes it, it sets the cpu_active
bit for this cpu. The cpu_active bit is the relevant one for the
scheduler to consider the cpu as a viable target.
With forced threaded interrupt handlers which imply forced threaded
softirqs we observed the following race:
cpu 0 cpu 1
bringup(cpu1);
set_cpu_online(smp_processor_id(), true);
local_irq_enable();
while (!cpu_online(cpu1));
timer_interrupt()
-> wake_up(softirq_thread_cpu1);
-> enqueue_on(softirq_thread_cpu1, cpu0);
^^^^
cpu_notify(CPU_ONLINE, cpu1);
-> sched_cpu_active(cpu1)
-> set_cpu_active((cpu1, true);
When an interrupt happens before the cpu_active bit is set by the cpu
which brought up the newly onlined cpu, then the scheduler refuses to
enqueue the woken thread which is bound to that newly onlined cpu on
that newly onlined cpu due to the not yet set cpu_active bit and
selects a fallback runqueue. Not really an expected and desirable
behaviour.
So far this has only been observed with forced hard/softirq threading,
but in theory this could happen without forced threaded hard/softirqs
as well. It's probably unobservable as it would take a massive
interrupt storm on the newly onlined cpu which causes the softirq loop
to wake up the softirq thread and an even longer delay of the cpu
which waits for the cpu_online bit.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Peter Zijlstra <peterz@infradead.org>
Cc: stable@kernel.org # 2.6.39
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86/amd-iommu: Fix boot crash with hidden PCI devices
x86/amd-iommu: Use only per-device dma_ops
x86/amd-iommu: Fix 3 possible endless loops
* 'kvm-updates/3.0' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: Initialize kvm before registering the mmu notifier
KVM: x86: use proper port value when checking io instruction permission
KVM: add missing void __user * cast to access_ok() call
