In recent stress tests, it was found that pvclock-based systems
could seriously warp in smp systems. Using ingo's time-warp-test.c,
I could trigger a scenario as bad as 1.5mi warps a minute in some systems.
(to be fair, it wasn't that bad in most of them). Investigating further, I
found out that such warps were caused by the very offset-based calculation
pvclock is based on.
This happens even on some machines that report constant_tsc in its tsc flags,
specially on multi-socket ones.
Two reads of the same kernel timestamp at approx the same time, will likely
have tsc timestamped in different occasions too. This means the delta we
calculate is unpredictable at best, and can probably be smaller in a cpu
that is legitimately reading clock in a forward ocasion.
Some adjustments on the host could make this window less likely to happen,
but still, it pretty much poses as an intrinsic problem of the mechanism.
A while ago, I though about using a shared variable anyway, to hold clock
last state, but gave up due to the high contention locking was likely
to introduce, possibly rendering the thing useless on big machines. I argue,
however, that locking is not necessary.
We do a read-and-return sequence in pvclock, and between read and return,
the global value can have changed. However, it can only have changed
by means of an addition of a positive value. So if we detected that our
clock timestamp is less than the current global, we know that we need to
return a higher one, even though it is not exactly the one we compared to.
OTOH, if we detect we're greater than the current time source, we atomically
replace the value with our new readings. This do causes contention on big
boxes (but big here means *BIG*), but it seems like a good trade off, since
it provide us with a time source guaranteed to be stable wrt time warps.
After this patch is applied, I don't see a single warp in time during 5 days
of execution, in any of the machines I saw them before.
Signed-off-by: Glauber Costa <glommer@redhat.com>
Acked-by: Zachary Amsden <zamsden@redhat.com>
CC: Jeremy Fitzhardinge <jeremy@goop.org>
CC: Avi Kivity <avi@redhat.com>
CC: Marcelo Tosatti <mtosatti@redhat.com>
CC: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch removes one padding byte and transform it into a flags
field. New versions of guests using pvclock will query these flags
upon each read.
Flags, however, will only be interpreted when the guest decides to.
It uses the pvclock_valid_flags function to signal that a specific
set of flags should be taken into consideration. Which flags are valid
are usually devised via HV negotiation.
Signed-off-by: Glauber Costa <glommer@redhat.com>
CC: Jeremy Fitzhardinge <jeremy@goop.org>
Acked-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch fixes a bug in the KVM efer-msr write path. If a
guest writes to a reserved efer bit the set_efer function
injects the #GP directly. The architecture dependent wrmsr
function does not see this, assumes success and advances the
rip. This results in a #GP in the guest with the wrong rip.
This patch fixes this by reporting efer write errors back to
the architectural wrmsr function.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch disables the possibility for a l2-guest to do a
VMMCALL directly into the host. This would happen if the
l1-hypervisor doesn't intercept VMMCALL and the l2-guest
executes this instruction.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The patch merged recently which allowed to mark an exception
as reinjected has a bug as it always marks the exception as
reinjected. This breaks nested-svm shadow-on-shadow
implementation.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Wallclock writing uses an unprotected global variable to hold the version;
this can cause one guest to interfere with another if both write their
wallclock at the same time.
Acked-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
On svm, kvm_read_pdptr() may require reading guest memory, which can sleep.
Push the spinlock into mmu_alloc_roots(), and only take it after we've read
the pdptr.
Tested-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Per document, for feature control MSR:
Bit 1 enables VMXON in SMX operation. If the bit is clear, execution
of VMXON in SMX operation causes a general-protection exception.
Bit 2 enables VMXON outside SMX operation. If the bit is clear, execution
of VMXON outside SMX operation causes a general-protection exception.
This patch is to enable this kind of check with SMX for VMXON in KVM.
Signed-off-by: Shane Wang <shane.wang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The recent changes to emulate string instructions without entering guest
mode exposed a bug where pending interrupts are not properly reflected
in ready_for_interrupt_injection.
The result is that userspace overwrites a previously queued interrupt,
when irqchip's are emulated in userspace.
Fix by always updating state before returning to userspace.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
When EPT is enabled, we cannot emulate EFER.NX=0 through the shadow page
tables. This causes accesses through ptes with bit 63 set to succeed instead
of failing a reserved bit check.
Signed-off-by: Avi Kivity <avi@redhat.com>
Some guest msr values cannot be used on the host (for example. EFER.NX=0),
so we need to switch them atomically during guest entry or exit.
Add a facility to program the vmx msr autoload registers accordingly.
Signed-off-by: Avi Kivity <avi@redhat.com>
vmx and svm vcpus have different contents and therefore may have different
alignmment requirements. Let each specify its required alignment.
Signed-off-by: Avi Kivity <avi@redhat.com>
Move unsync/sync tracepoints to the proper place, it's good
for us to obtain unsync page live time
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If the guest is 32-bit, we should use 'quadrant' to adjust gpa
offset
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
kvm_mmu_remove_one_alloc_mmu_page() assumes kvm_mmu_zap_page() only reclaims
only one sp, but that's not the case. This will cause mmu shrinker returns
a wrong number. This patch fix the counting error.
Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
For TDP mode, avoid creating multiple page table roots for the single
guest-to-host physical address map by fixing the inputs used for the
shadow page table hash in mmu_alloc_roots().
Signed-off-by: Eric Northup <digitaleric@google.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch prevents MCE intercepts from being propagated
into the L1 guest if they happened in an L2 guest.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds logic to kvm/x86 which allows to mark an
injected exception as reinjected. This allows to remove an
ugly hack from svm_complete_interrupts that prevented
exceptions from being reinjected at all in the nested case.
The hack was necessary because an reinjected exception into
the nested guest could cause a nested vmexit emulation. But
reinjected exceptions must not intercept. The downside of
the hack is that a exception that in injected could get
lost.
This patch fixes the problem and puts the code for it into
generic x86 files because. Nested-VMX will likely have the
same problem and could reuse the code.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch implements the reporting of the emulated SVM
features to userspace instead of the real hardware
capabilities. Every real hardware capability needs emulation
in nested svm so the old behavior was broken.
Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds the get_supported_cpuid callback to
kvm_x86_ops. It will be used in do_cpuid_ent to delegate the
decission about some supported cpuid bits to the
architecture modules.
Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch implements propagation of a failes guest vmrun
back into the guest instead of killing the whole guest.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch syncs cr0 and cr3 from the vmcb to the kvm state
before nested intercept handling is done. This allows to
simplify the vmexit path.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch fixes a bug where a nested guest always went over
the same instruction because the rip was not advanced on a
nested vmexit.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The patch introducing nested nmi handling had a bug. The
check does not belong to enable_nmi_window but must be in
nmi_allowed. This patch fixes this.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Fix bug of the exception path, free allocated vpid when fail
to create vcpu.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Since gfn is not changed in the for loop, we do not need to call
gfn_to_memslot_unaliased() under the loop, and it is safe to move
it out.
Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Make use of is_large_pte() instead of checking PT_PAGE_SIZE_MASK
bit directly.
Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Move first pte address calculation out of loop to save some cycles.
Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Since commit bf47a760f6, we no longer handle ptes with the global bit
set specially, so there is no reason to distinguish between shadow pages
created with cr4.gpe set and clear.
Such tracking is expensive when the guest toggles cr4.pge, so drop it.
Signed-off-by: Avi Kivity <avi@redhat.com>
Quote from Avi:
|Just change the assignment to a 'goto restart;' please,
|I don't like playing with list_for_each internals.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If kvm_task_switch() fails code exits to userspace without specifying
exit reason, so the previous exit reason is reused by userspace. Fix
this by specifying exit reason correctly.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
emulator_task_switch() should return -1 for failure and 0 for success to
the caller, just like x86_emulate_insn() does.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
There is no real distinction between glevels=3 and glevels=4; both have
exactly the same format and the code is treated exactly the same way. Drop
role.glevels and replace is with role.cr4_pae (which is meaningful). This
simplifies the code a bit.
As a side effect, it allows sharing shadow page tables between pae and
longmode guest page tables at the same guest page.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
When a fault triggers a task switch, the error code, if existent, has to
be pushed on the new task's stack. Implement the missing bits.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Stop the switch immediately if task_switch_16/32 returned an error. Only
if that step succeeded, the switch should actually take place and update
any register states.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
We can call kvm_mmu_pte_write() directly from
emulator_cmpxchg_emulated() instead of passing mmu_only down to
emulator_write_emulated_onepage() and call it there.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Currently both SVM and VMX have their own DR handling code. Move it to
x86.c.
Acked-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Glauber Costa
Roedel, Joerg
Avi Kivity
Shane Wang
Marcelo Tosatti
Xiao Guangrong
Gui Jianfeng
Eric Northup
Lai Jiangshan
Wei Yongjun
Gleb Natapov
Jan Kiszka