Under Xen 64 bit guests actually run their kernel in ring 3,
however the hypervisor takes care of squashing descriptor the
RPLs transparently (in order to allow them to continue to
differentiate between user and kernel space CS using the RPL).
Therefore the Xen paravirt backend should use RPL==0 instead of
1 (or 3). Using RPL==1 causes generic arch code to take
incorrect code paths because it uses "testl $3, <CS>, je foo"
type tests for a userspace CS and this considers 1==userspace.
This issue was previously masked because get_kernel_rpl() was
omitted when setting CS in kernel_thread(). This was fixed when
kernel_thread() was unified with 32 bit in
f443ff4201.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Christian Kujau <lists@nerdbynature.de>
Cc: Jeremy Fitzhardinge <Jeremy.Fitzhardinge@citrix.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Brian Gerst <brgerst@gmail.com>
LKML-Reference: <1263377768-19600-2-git-send-email-ian.campbell@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The raw_spin* namespace was taken by lockdep for the architecture
specific implementations. raw_spin_* would be the ideal name space for
the spinlocks which are not converted to sleeping locks in preempt-rt.
Linus suggested to convert the raw_ to arch_ locks and cleanup the
name space instead of using an artifical name like core_spin,
atomic_spin or whatever
No functional change.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: linux-arch@vger.kernel.org
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu: (34 commits)
m68k: rename global variable vmalloc_end to m68k_vmalloc_end
percpu: add missing per_cpu_ptr_to_phys() definition for UP
percpu: Fix kdump failure if booted with percpu_alloc=page
percpu: make misc percpu symbols unique
percpu: make percpu symbols in ia64 unique
percpu: make percpu symbols in powerpc unique
percpu: make percpu symbols in x86 unique
percpu: make percpu symbols in xen unique
percpu: make percpu symbols in cpufreq unique
percpu: make percpu symbols in oprofile unique
percpu: make percpu symbols in tracer unique
percpu: make percpu symbols under kernel/ and mm/ unique
percpu: remove some sparse warnings
percpu: make alloc_percpu() handle array types
vmalloc: fix use of non-existent percpu variable in put_cpu_var()
this_cpu: Use this_cpu_xx in trace_functions_graph.c
this_cpu: Use this_cpu_xx for ftrace
this_cpu: Use this_cpu_xx in nmi handling
this_cpu: Use this_cpu operations in RCU
this_cpu: Use this_cpu ops for VM statistics
...
Fix up trivial (famous last words) global per-cpu naming conflicts in
arch/x86/kvm/svm.c
mm/slab.c
* 'bugfix' of git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen:
xen: try harder to balloon up under memory pressure.
Xen balloon: fix totalram_pages counting.
xen: explicitly create/destroy stop_machine workqueues outside suspend/resume region.
xen: improve error handling in do_suspend.
xen: don't leak IRQs over suspend/resume.
xen: call clock resume notifier on all CPUs
xen: use iret for return from 64b kernel to 32b usermode
xen: don't call dpm_resume_noirq() with interrupts disabled.
xen: register runstate info for boot CPU early
xen: register runstate on secondary CPUs
xen: register timer interrupt with IRQF_TIMER
xen: correctly restore pfn_to_mfn_list_list after resume
xen: restore runstate_info even if !have_vcpu_info_placement
xen: re-register runstate area earlier on resume.
xen: wait up to 5 minutes for device connetion
xen: improvement to wait_for_devices()
xen: fix is_disconnected_device/exists_disconnected_device
xen/xenbus: make DEVICE_ATTR()s static
* 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (36 commits)
x86, mm: Correct the implementation of is_untracked_pat_range()
x86/pat: Trivial: don't create debugfs for memtype if pat is disabled
x86, mtrr: Fix sorting of mtrr after subtracting
x86: Move find_smp_config() earlier and avoid bootmem usage
x86, platform: Change is_untracked_pat_range() to bool; cleanup init
x86: Change is_ISA_range() into an inline function
x86, mm: is_untracked_pat_range() takes a normal semiclosed range
x86, mm: Call is_untracked_pat_range() rather than is_ISA_range()
x86: UV SGI: Don't track GRU space in PAT
x86: SGI UV: Fix BAU initialization
x86, numa: Use near(er) online node instead of roundrobin for NUMA
x86, numa, bootmem: Only free bootmem on NUMA failure path
x86: Change crash kernel to reserve via reserve_early()
x86: Eliminate redundant/contradicting cache line size config options
x86: When cleaning MTRRs, do not fold WP into UC
x86: remove "extern" from function prototypes in <asm/proto.h>
x86, mm: Report state of NX protections during boot
x86, mm: Clean up and simplify NX enablement
x86, pageattr: Make set_memory_(x|nx) aware of NX support
x86, sleep: Always save the value of EFER
...
Fix up conflicts (added both iommu_shutdown and is_untracked_pat_range)
to 'struct x86_platform_ops') in
arch/x86/include/asm/x86_init.h
arch/x86/kernel/x86_init.c
Commit ae21ee65e8 "PCI: acs p2p upsteram
forwarding enabling" doesn't actually enable ACS.
Add a function to pci core to allow an IOMMU to request that ACS
be enabled. The existing mechanism of using iommu_found() in the pci
core to know when ACS should be enabled doesn't actually work due to
initialization order; iommu has only been detected not initialized.
Have Intel and AMD IOMMUs request ACS, and Xen does as well during early
init of dom0.
Cc: Allen Kay <allen.m.kay@intel.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
tick_resume() is never called on secondary processors. Presumably this
is because they are offlined for suspend on native and so this is
normally taken care of in the CPU onlining path. Under Xen we keep all
CPUs online over a suspend.
This patch papers over the issue for me but I will investigate a more
generic, less hacky, way of doing to the same.
tick_suspend is also only called on the boot CPU which I presume should
be fixed too.
Signed-off-by: Ian Campbell <Ian.Campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
If Xen wants to return to a 32b usermode with sysret it must use the
right form. When using VCGF_in_syscall to trigger this, it looks at
the code segment and does a 32b sysret if it is FLAT_USER_CS32.
However, this is different from __USER32_CS, so it fails to return
properly if we use the normal Linux segment.
So avoid the whole mess by dropping VCGF_in_syscall and simply use
plain iret to return to usermode.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: Jan Beulich <jbeulich@novell.com>
Cc: Stable Kernel <stable@kernel.org>
printk timestamping uses sched_clock, which in turn relies on runstate
info under Xen. So make sure we set it up before any printks can
be called.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
The commit "xen: re-register runstate area earlier on resume" caused us
to never try and setup the runstate area for secondary CPUs. Ensure that
we do this...
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
Otherwise the timer is disabled by dpm_suspend_noirq() which in turn prevents
correct operation of stop_machine on multi-processor systems and breaks
suspend.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
pvops kernels >= 2.6.30 can currently only be saved and restored once. The
second attempt to save results in:
ERROR Internal error: Frame# in pfn-to-mfn frame list is not in pseudophys
ERROR Internal error: entry 0: p2m_frame_list[0] is 0xf2c2c2c2, max 0x120000
ERROR Internal error: Failed to map/save the p2m frame list
I finally narrowed it down to:
commit cdaead6b4e
Author: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Date: Fri Feb 27 15:34:59 2009 -0800
xen: split construction of p2m mfn tables from registration
Build the p2m_mfn_list_list early with the rest of the p2m table, but
register it later when the real shared_info structure is in place.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
The unforeseen side-effect of this change was to cause the mfn list list to not
be rebuilt on resume. Prior to this change it would have been rebuilt via
xen_post_suspend() -> xen_setup_shared_info() -> xen_setup_mfn_list_list().
Fix by explicitly calling xen_build_mfn_list_list() from xen_post_suspend().
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
Even if have_vcpu_info_placement is not set, we still need to set up
the runstate area on each resumed vcpu.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
This is necessary to ensure the runstate area is available to
xen_sched_clock before any calls to printk which will require it in
order to provide a timestamp.
I chose to pull the xen_setup_runstate_info out of xen_time_init into
the caller in order to maintain parity with calling
xen_setup_runstate_info separately from calling xen_time_resume.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
The 32- and 64-bit code used very different mechanisms for enabling
NX, but even the 32-bit code was enabling NX in head_32.S if it is
available. Furthermore, we had a bewildering collection of tests for
the available of NX.
This patch:
a) merges the 32-bit set_nx() and the 64-bit check_efer() function
into a single x86_configure_nx() function. EFER control is left
to the head code.
b) eliminates the nx_enabled variable entirely. Things that need to
test for NX enablement can verify __supported_pte_mask directly,
and cpu_has_nx gives the supported status of NX.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Vegard Nossum <vegardno@ifi.uio.no>
Cc: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Chris Wright <chrisw@sous-sol.org>
LKML-Reference: <1258154897-6770-5-git-send-email-hpa@zytor.com>
Acked-by: Kees Cook <kees.cook@canonical.com>
Move xen_domain and related tests out of asm-x86 to xen/xen.h so they
can be included whenever they are necessary.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
A Xen guest never needs to know about extended topology, and knowing
would just confuse it.
This patch just zeros ebx in leaf 0xb which indicates no topology info,
preventing a crash under Xen on cpus which support this leaf.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
This patch updates percpu related symbols in xen such that percpu
symbols are unique and don't clash with local symbols. This serves
two purposes of decreasing the possibility of global percpu symbol
collision and allowing dropping per_cpu__ prefix from percpu symbols.
* arch/x86/xen/smp.c, arch/x86/xen/time.c, arch/ia64/xen/irq_xen.c:
add xen_ prefix to percpu variables
* arch/ia64/xen/time.c: add xen_ prefix to percpu variables, drop
processed_ prefix and make them static
Partly based on Rusty Russell's "alloc_percpu: rename percpu vars
which cause name clashes" patch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Chris Wright <chrisw@sous-sol.org>
xen_setup_stackprotector() ends up trying to set page protections,
so we need to have vm_mmu_ops set up before trying to do so.
Failing to do so causes an early boot crash.
[ Impact: Fix early crash under Xen. ]
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: Remove redundant non-NUMA topology functions
x86: early_printk: Protect against using the same device twice
x86: Reduce verbosity of "PAT enabled" kernel message
x86: Reduce verbosity of "TSC is reliable" message
x86: mce: Use safer ways to access MCE registers
x86: mce, inject: Use real inject-msg in raise_local
x86: mce: Fix thermal throttling message storm
x86: mce: Clean up thermal throttling state tracking code
x86: split NX setup into separate file to limit unstack-protected code
xen: check EFER for NX before setting up GDT mapping
x86: Cleanup linker script using new linker script macros.
x86: Use section .data.page_aligned for the idt_table.
x86: convert to use __HEAD and HEAD_TEXT macros.
x86: convert compressed loader to use __HEAD and HEAD_TEXT macros.
x86: fix fragile computation of vsyscall address
Makes code futureproof against the impending change to mm->cpu_vm_mask (to be a pointer).
It's also a chance to use the new cpumask_ ops which take a pointer
(the older ones are deprecated, but there's no hurry for arch code).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
x86-64 assumes NX is available by default, so we need to
explicitly check for it before using NX. Some first-generation
Intel x86-64 processors didn't support NX, and even recent systems
allow it to be disabled in BIOS.
[ Impact: prevent Xen crash on NX-less 64-bit machines ]
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stable Kernel <stable@kernel.org>
* 'x86-platform-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (38 commits)
x86: Move get/set_wallclock to x86_platform_ops
x86: platform: Fix section annotations
x86: apic namespace cleanup
x86: Distangle ioapic and i8259
x86: Add Moorestown early detection
x86: Add hardware_subarch ID for Moorestown
x86: Add early platform detection
x86: Move tsc_init to late_time_init
x86: Move tsc_calibration to x86_init_ops
x86: Replace the now identical time_32/64.c by time.c
x86: time_32/64.c unify profile_pc
x86: Move calibrate_cpu to tsc.c
x86: Make timer setup and global variables the same in time_32/64.c
x86: Remove mca bus ifdef from timer interrupt
x86: Simplify timer_ack magic in time_32.c
x86: Prepare unification of time_32/64.c
x86: Remove do_timer hook
x86: Add timer_init to x86_init_ops
x86: Move percpu clockevents setup to x86_init_ops
x86: Move xen_post_allocator_init into xen_pagetable_setup_done
...
Fix up conflicts in arch/x86/include/asm/io_apic.h
get/set_wallclock() have already a set of platform dependent
implementations (default, EFI, paravirt). MRST will add another
variant.
Moving them to platform ops simplifies the existing code and minimizes
the effort to integrate new variants.
Signed-off-by: Feng Tang <feng.tang@intel.com>
LKML-Reference: <new-submission>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
* 'x86-xen-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: split __phys_addr out into separate file
xen: use stronger barrier after unlocking lock
xen: only enable interrupts while actually blocking for spinlock
xen: make -fstack-protector work under Xen
* 'x86-cpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (22 commits)
x86: Fix code patching for paravirt-alternatives on 486
x86, msr: change msr-reg.o to obj-y, and export its symbols
x86: Use hard_smp_processor_id() to get apic id for AMD K8 cpus
x86, sched: Workaround broken sched domain creation for AMD Magny-Cours
x86, mcheck: Use correct cpumask for shared bank4
x86, cacheinfo: Fixup L3 cache information for AMD multi-node processors
x86: Fix CPU llc_shared_map information for AMD Magny-Cours
x86, msr: Fix msr-reg.S compilation with gas 2.16.1, on 32-bit too
x86: Move kernel_fpu_using to irq_fpu_usable in asm/i387.h
x86, msr: fix msr-reg.S compilation with gas 2.16.1
x86, msr: Export the register-setting MSR functions via /dev/*/msr
x86, msr: Create _on_cpu helpers for {rw,wr}msr_safe_regs()
x86, msr: Have the _safe MSR functions return -EIO, not -EFAULT
x86, msr: CFI annotations, cleanups for msr-reg.S
x86, asm: Make _ASM_EXTABLE() usable from assembly code
x86, asm: Add 32-bit versions of the combined CFI macros
x86, AMD: Disable wrongly set X86_FEATURE_LAHF_LM CPUID bit
x86, msr: Rewrite AMD rd/wrmsr variants
x86, msr: Add rd/wrmsr interfaces with preset registers
x86: add specific support for Intel Atom architecture
...
We need to have a stronger barrier between releasing the lock and
checking for any waiting spinners. A compiler barrier is not sufficient
because the CPU's ordering rules do not prevent the read xl->spinners
from happening before the unlock assignment, as they are different
memory locations.
We need to have an explicit barrier to enforce the write-read ordering
to different memory locations.
Because of it, I can't bring up > 4 HVM guests on one SMP machine.
[ Code and commit comments expanded -J ]
[ Impact: avoid deadlock when using Xen PV spinlocks ]
Signed-off-by: Yang Xiaowei <xiaowei.yang@intel.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Where possible we enable interrupts while waiting for a spinlock to
become free, in order to reduce big latency spikes in interrupt handling.
However, at present if we manage to pick up the spinlock just before
blocking, we'll end up holding the lock with interrupts enabled for a
while. This will cause a deadlock if we recieve an interrupt in that
window, and the interrupt handler tries to take the lock too.
Solve this by shrinking the interrupt-enabled region to just around the
blocking call.
[ Impact: avoid race/deadlock when using Xen PV spinlocks ]
Reported-by: "Yang, Xiaowei" <xiaowei.yang@intel.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
-fstack-protector uses a special per-cpu "stack canary" value.
gcc generates special code in each function to test the canary to make
sure that the function's stack hasn't been overrun.
On x86-64, this is simply an offset of %gs, which is the usual per-cpu
base segment register, so setting it up simply requires loading %gs's
base as normal.
On i386, the stack protector segment is %gs (rather than the usual kernel
percpu %fs segment register). This requires setting up the full kernel
GDT and then loading %gs accordingly. We also need to make sure %gs is
initialized when bringing up secondary cpus too.
To keep things consistent, we do the full GDT/segment register setup on
both architectures.
Because we need to avoid -fstack-protected code before setting up the GDT
and because there's no way to disable it on a per-function basis, several
files need to have stack-protector inhibited.
[ Impact: allow Xen booting with stack-protector enabled ]
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
For some reason, the _safe MSR functions returned -EFAULT, not -EIO.
However, the only user which cares about the return code as anything
other than a boolean is the MSR driver, which wants -EIO. Change it
to -EIO across the board.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
TSC calibration is modified by the vmware hypervisor and paravirt by
separate means. Moorestown wants to add its own calibration routine as
well. So make calibrate_tsc a proper x86_init_ops function and
override it by paravirt or by the early setup of the vmware
hypervisor.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The timer init code is convoluted with several quirks and the paravirt
timer chooser. Figuring out which code path is actually taken is not
for the faint hearted.
Move the numaq TSC quirk to tsc_pre_init x86_init_ops function and
replace the paravirt time chooser and the remaining x86 quirk with a
simple x86_init_ops function.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
paravirt overrides the setup of the default apic timers as per cpu
timers. Moorestown needs to override that as well.
Move it to x86_init_ops setup and create a separate x86_cpuinit struct
which holds the function for the secondary evtl. hotplugabble CPUs.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
We really do not need two paravirt/x86_init_ops functions which are
called in two consecutive source lines. Move the only user of
post_allocator_init into the already existing pagetable_setup_done
function.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Replace another obscure paravirt magic and move it to
x86_init_ops. Such a hook is also useful for embedded and special
hardware.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
ARCH_SETUP is a horrible leftover from the old arch/i386 mach support
code. It still has a lonely user in xen. Move it to x86_init_ops.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
irq_init is overridden by x86_quirks and by paravirts. Unify the whole
mess and make it an unconditional x86_init_ops function which defaults
to the standard function and can be overridden by the early platform
code.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
memory_setup is overridden by x86_quirks and by paravirts with weak
functions and quirks. Unify the whole mess and make it an
unconditional x86_init_ops function which defaults to the standard
function and can be overridden by the early platform code.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Initialize cx before calling xen_cpuid(), in order to suppress the
"may be used uninitialized in this function" warning.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Xen always runs on CPUs which properly support WP enforcement in
privileged mode, so there's no need to test for it.
This also works around a crash reported by Arnd Hannemann, though I
think its just a band-aid for that case.
Reported-by: Arnd Hannemann <hannemann@nets.rwth-aachen.de>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make sure the stack-protector segment registers are properly set up
before calling any functions which may have stack-protection compiled
into them.
[ Impact: prevent Xen early-boot crash when stack-protector is enabled ]
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
* 'x86-xen-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (42 commits)
xen: cache cr0 value to avoid trap'n'emulate for read_cr0
xen/x86-64: clean up warnings about IST-using traps
xen/x86-64: fix breakpoints and hardware watchpoints
xen: reserve Xen start_info rather than e820 reserving
xen: add FIX_TEXT_POKE to fixmap
lguest: update lazy mmu changes to match lguest's use of kvm hypercalls
xen: honour VCPU availability on boot
xen: add "capabilities" file
xen: drop kexec bits from /sys/hypervisor since kexec isn't implemented yet
xen/sys/hypervisor: change writable_pt to features
xen: add /sys/hypervisor support
xen/xenbus: export xenbus_dev_changed
xen: use device model for suspending xenbus devices
xen: remove suspend_cancel hook
xen/dev-evtchn: clean up locking in evtchn
xen: export ioctl headers to userspace
xen: add /dev/xen/evtchn driver
xen: add irq_from_evtchn
xen: clean up gate trap/interrupt constants
xen: set _PAGE_NX in __supported_pte_mask before pagetable construction
...
Xiaohui Xin and some other folks at Intel have been looking into what's
behind the performance hit of paravirt_ops when running native.
It appears that the hit is entirely due to the paravirtualized
spinlocks introduced by:
| commit 8efcbab674
| Date: Mon Jul 7 12:07:51 2008 -0700
|
| paravirt: introduce a "lock-byte" spinlock implementation
The extra call/return in the spinlock path is somehow
causing an increase in the cycles/instruction of somewhere around 2-7%
(seems to vary quite a lot from test to test). The working theory is
that the CPU's pipeline is getting upset about the
call->call->locked-op->return->return, and seems to be failing to
speculate (though I haven't seen anything definitive about the precise
reasons). This doesn't entirely make sense, because the performance
hit is also visible on unlock and other operations which don't involve
locked instructions. But spinlock operations clearly swamp all the
other pvops operations, even though I can't imagine that they're
nearly as common (there's only a .05% increase in instructions
executed).
If I disable just the pv-spinlock calls, my tests show that pvops is
identical to non-pvops performance on native (my measurements show that
it is actually about .1% faster, but Xiaohui shows a .05% slowdown).
Summary of results, averaging 10 runs of the "mmperf" test, using a
no-pvops build as baseline:
nopv Pv-nospin Pv-spin
CPU cycles 100.00% 99.89% 102.18%
instructions 100.00% 100.10% 100.15%
CPI 100.00% 99.79% 102.03%
cache ref 100.00% 100.84% 100.28%
cache miss 100.00% 90.47% 88.56%
cache miss rate 100.00% 89.72% 88.31%
branches 100.00% 99.93% 100.04%
branch miss 100.00% 103.66% 107.72%
branch miss rt 100.00% 103.73% 107.67%
wallclock 100.00% 99.90% 102.20%
The clear effect here is that the 2% increase in CPI is
directly reflected in the final wallclock time.
(The other interesting effect is that the more ops are
out of line calls via pvops, the lower the cache access
and miss rates. Not too surprising, but it suggests that
the non-pvops kernel is over-inlined. On the flipside,
the branch misses go up correspondingly...)
So, what's the fix?
Paravirt patching turns all the pvops calls into direct calls, so
_spin_lock etc do end up having direct calls. For example, the compiler
generated code for paravirtualized _spin_lock is:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq *0xffffffff805a5b30
<_spin_lock+22>: retq
The indirect call will get patched to:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq <__ticket_spin_lock>
<_spin_lock+20>: nop; nop /* or whatever 2-byte nop */
<_spin_lock+22>: retq
One possibility is to inline _spin_lock, etc, when building an
optimised kernel (ie, when there's no spinlock/preempt
instrumentation/debugging enabled). That will remove the outer
call/return pair, returning the instruction stream to a single
call/return, which will presumably execute the same as the non-pvops
case. The downsides arel 1) it will replicate the
preempt_disable/enable code at eack lock/unlock callsite; this code is
fairly small, but not nothing; and 2) the spinlock definitions are
already a very heavily tangled mass of #ifdefs and other preprocessor
magic, and making any changes will be non-trivial.
The other obvious answer is to disable pv-spinlocks. Making them a
separate config option is fairly easy, and it would be trivial to
enable them only when Xen is enabled (as the only non-default user).
But it doesn't really address the common case of a distro build which
is going to have Xen support enabled, and leaves the open question of
whether the native performance cost of pv-spinlocks is worth the
performance improvement on a loaded Xen system (10% saving of overall
system CPU when guests block rather than spin). Still it is a
reasonable short-term workaround.
[ Impact: fix pvops performance regression when running native ]
Analysed-by: "Xin Xiaohui" <xiaohui.xin@intel.com>
Analysed-by: "Li Xin" <xin.li@intel.com>
Analysed-by: "Nakajima Jun" <jun.nakajima@intel.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Xen-devel <xen-devel@lists.xensource.com>
LKML-Reference: <4A0B62F7.5030802@goop.org>
[ fixed the help text ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>