Recent enhancement of rb-tree based lookup exposed a bug with the lookup
mechanism in the reserve_memtype() which ensures that there are no conflicting
memtype requests for the memory range.
memtype_rb_search() returns an entry which has a start address <= new start
address. And from here we traverse the linear linked list to check if there
any conflicts with the existing mappings. As the rbtree is based on the
start address of the memory range, it is quite possible that we have several
overlapped mappings whose start address is much less than new requested start
but the end is >= new requested end. This results in conflicting memtype
mappings.
Same bug exists with the old code which uses cached_entry from where
we traverse the linear linked list. But the new rb-tree code exposes this
bug fairly easily.
For now, don't use the memtype_rb_search() and always start the search from
the head of linear linked list in reserve_memtype(). Linear linked list
for most of the systems grow's to few 10's of entries(as we track memory type
of RAM pages using struct page). So we should be ok for now.
We still retain the rbtree and use it to speed up free_memtype() which
doesn't have the same bug(as we know what exactly we are searching for
in free_memtype).
Also use list_for_each_entry_from() in free_memtype() so that we start
the search from rb-tree lookup result.
Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
LKML-Reference: <1253136483.4119.12.camel@sbs-t61.sc.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
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>
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>
Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
init_IRQ() and x86_late_time_init() are missing __init annotations.
The x86 platform ops variables are annotated, but the annotation needs
to be put between the variable name and the "=" of the initializer.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
APERF/MPERF support for cpu_power.
APERF/MPERF is arch defined to be a relative scale of work capacity
per logical cpu, this is assumed to include SMT and Turbo mode.
APERF/MPERF are specified to both reset to 0 when either counter
wraps, which is highly inconvenient, since that'll give a blimp
when that happens. The manual specifies writing 0 to the counters
after each read, but that's 1) too expensive, and 2) destroys the
possibility of sharing these counters with other users, so we live
with the blimp - the other existing user does too.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Move some of the aperf/mperf code out from the cpufreq driver
thingy so that other people can enjoy it too.
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Yanmin <yanmin_zhang@linux.intel.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Len Brown <len.brown@intel.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: cpufreq@vger.kernel.org
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Move the APERFMPERF capacility into a X86_FEATURE flag so that it
can be used outside of the acpi cpufreq driver.
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Yanmin <yanmin_zhang@linux.intel.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Len Brown <len.brown@intel.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: cpufreq@vger.kernel.org
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
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>
Fix compilation error in arch/x86/kernel/cpu/mcheck/mce-severity.c
when CONFIG_DEBUG_FS is disabled, introduced in commit
5be9ed251f.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Move NB decoder along with required defines to EDAC MCE core. Add
registration routines for further decoding of the MCE info in the AMD64
EDAC module.
CC: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Only 24 bytes needs to be reserved on the stack for the function graph
tracer on x86_64.
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
LKML-Reference: <20090729085837.GB4998@jolsa.lab.eng.brq.redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently we are not including randomized stack size when calculating
mmap_base address in arch_pick_mmap_layout for topdown case. This might
cause that mmap_base starts in the stack reserved area because stack is
randomized by 1GB for 64b (8MB for 32b) and the minimum gap is 128MB.
If the stack really grows down to mmap_base then we can get silent mmap
region overwrite by the stack values.
Let's include maximum stack randomization size into MIN_GAP which is
used as the low bound for the gap in mmap.
Signed-off-by: Michal Hocko <mhocko@suse.cz>
LKML-Reference: <1252400515-6866-1-git-send-email-mhocko@suse.cz>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Stable Team <stable@kernel.org>
As reported in <http://bugs.debian.org/511703> and
<http://bugs.debian.org/515982>, kernels with paravirt-alternatives
enabled crash in text_poke_early() on at least some 486-class
processors.
The problem is that text_poke_early() itself uses inline functions
affected by paravirt-alternatives and so will modify instructions that
have already been prefetched. Pentium and later processors will
invalidate the prefetched instructions in this case, but 486-class
processors do not.
Change sync_core() to limit prefetching on 486-class (and 386-class)
processors, and move the call to sync_core() above the call to the
modifiable local_irq_restore().
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
LKML-Reference: <1252547631.3423.134.camel@localhost>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The dynamic function tracer relys on the macro P6_NOP5 always being
an atomic NOP. If for some reason it is changed to be two operations
(like a nop2 nop3) it can faults within the kernel when the function
tracer modifies the code.
This patch adds a comment to note that the P6_NOPs are expected to
be atomic. This will hopefully prevent anyone from changing that.
Reported-by: Mathieu Desnoyer <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Split __phys_addr out into its own file so we can disable
-fstack-protector in a fine-grained fashion. Also it doesn't
have terribly much to do with the rest of ioremap.c.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Debug registers may only be accessed from cpl 0. Unfortunately, vmx will
code to emulate the instruction even though it was issued from guest
userspace, possibly leading to an unexpected trap later.
Cc: stable@kernel.org
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
No need to call it before each kvm_(set|get)_msr_common()
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Only reload debug register 6 if we're running with the guest's
debug registers. Saves around 150 cycles from the guest lightweight
exit path.
dr6 contains a couple of bits that are updated on #DB, so intercept
that unconditionally and update those bits then.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Instead of saving the debug registers from the processor to a kvm data
structure, rely in the debug registers stored in the thread structure.
This allows us not to save dr6 and dr7.
Reduces lightweight vmexit cost by 350 cycles, or 11 percent.
Signed-off-by: Avi Kivity <avi@redhat.com>
Commit b8bcfe997e made paravirt pte updates synchronous in interrupt
context.
Unfortunately the KVM pv mmu code caches the lazy/nonlazy mode
internally, so a pte update from interrupt context during a lazy mmu
operation can be batched while it should be performed synchronously.
https://bugzilla.redhat.com/show_bug.cgi?id=518022
Drop the internal mode variable and use paravirt_get_lazy_mode(), which
returns the correct state.
Cc: stable@kernel.org
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The use of __pa() to calculate the address of a C-visible symbol
is wrong, and can lead to unpredictable results. See arch/x86/include/asm/page.h
for details.
It should be replaced with __pa_symbol(), that does the correct math here,
by taking relocations into account. This ensures the correct wallclock data
structure physical address is passed to the hypervisor.
Cc: stable@kernel.org
Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Don't call adjust_vmx_controls() two times for the same control.
It restores options that were dropped earlier. This loses us the cr8
exit control, which causes a massive performance regression Windows x64.
Cc: stable@kernel.org
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
We know no pages are protected, so we can short-circuit the whole thing
(including fairly nasty guest memory accesses).
Signed-off-by: Avi Kivity <avi@redhat.com>
QNX update WP bit when paging enabled, which is not covered yet. This one fix
QNX boot with EPT.
Cc: stable@kernel.org
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Segment descriptors tables can be placed on two non-contiguous pages.
This patch makes reading segment descriptors by linear address.
Signed-off-by: Mikhail Ershov <Mike.Ershov@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Add missing decoder flags for adc and sbb instructions
(opcodes 0x14-0x15, 0x1c-0x1d)
Signed-off-by: Mohammed Gamal <m.gamal005@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
According to 16.2.5 in the SDM, eflags.vm in the tss is consulted before loading
and new segments. If eflags.vm == 1, then the segments are treated as 16-bit
segments. The LDTR and TR are not normally available in vm86 mode so if they
happen to somehow get loaded, they need to be treated as 32-bit segments.
This fixes an invalid vmentry failure in a custom OS that was happening after
a task switch into vm8086 mode. Since the segments were being mistakenly
treated as 32-bit, we loaded garbage state.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
We set rflags.vm86 when virtualizing real mode to do through vm8086 mode;
so we need to take it out again when reading rflags.
Signed-off-by: Avi Kivity <avi@redhat.com>
Since on vcpu entry we do it only if apic is enabled we should do
it when TPR is changed while apic is disabled. This happens when windows
resets HW without setting TPR to zero.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Nested SVM is (in my experience) stable enough to be enabled by
default. So omit the requirement to pass a module parameter.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Not checking for this flag breaks any nested hypervisor that does not
set VINTR. So fix it with this patch.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch removes one indentation level from nested_svm_intr and
makes the logic more readable.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This check is not necessary. We have to sync the vcpu->arch.cr2 always
back to the VMCB. This patch remove the is_nested check.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch moves the handling for special nested vmexits like #pf to a
separate function. This makes the kvm_override parameter obsolete and
makes the code more readable.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
If nested svm fails to load the msrpm the vmrun succeeds with the old
msrpm which is not correct. This patch changes the logic to roll back
to host mode in case the msrpm cannot be loaded.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch removes the usage of nested_svm_do from the vmrun emulation
path.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Suresh Siddha