The async page fault injection into kernel space creates more problems than
it solves. The host has absolutely no knowledge about the state of the
guest if the fault happens in CPL0. The only restriction for the host is
interrupt disabled state. If interrupts are enabled in the guest then the
exception can hit arbitrary code. The HALT based wait in non-preemotible
code is a hacky replacement for a proper hypercall.
For the ongoing work to restrict instrumentation and make the RCU idle
interaction well defined the required extra work for supporting async
pagefault in CPL0 is just not justified and creates complexity for a
dubious benefit.
The CPL3 injection is well defined and does not cause any issues as it is
more or less the same as a regular page fault from CPL3.
Suggested-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Alexandre Chartre <alexandre.chartre@oracle.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200505134059.369802541@linutronix.de
While working on the entry consolidation I stumbled over the KVM async page
fault handler and kvm_async_pf_task_wait() in particular. It took me a
while to realize that the randomly sprinkled around rcu_irq_enter()/exit()
invocations are just cargo cult programming. Several patches "fixed" RCU
splats by curing the symptoms without noticing that the code is flawed
from a design perspective.
The main problem is that this async injection is not based on a proper
handshake mechanism and only respects the minimal requirement, i.e. the
guest is not in a state where it has interrupts disabled.
Aside of that the actual code is a convoluted one fits it all swiss army
knife. It is invoked from different places with different RCU constraints:
1) Host side:
vcpu_enter_guest()
kvm_x86_ops->handle_exit()
kvm_handle_page_fault()
kvm_async_pf_task_wait()
The invocation happens from fully preemptible context.
2) Guest side:
The async page fault interrupted:
a) user space
b) preemptible kernel code which is not in a RCU read side
critical section
c) non-preemtible kernel code or a RCU read side critical section
or kernel code with CONFIG_PREEMPTION=n which allows not to
differentiate between #2b and #2c.
RCU is watching for:
#1 The vCPU exited and current is definitely not the idle task
#2a The #PF entry code on the guest went through enter_from_user_mode()
which reactivates RCU
#2b There is no preemptible, interrupts enabled code in the kernel
which can run with RCU looking away. (The idle task is always
non preemptible).
I.e. all schedulable states (#1, #2a, #2b) do not need any of this RCU
voodoo at all.
In #2c RCU is eventually not watching, but as that state cannot schedule
anyway there is no point to worry about it so it has to invoke
rcu_irq_enter() before running that code. This can be optimized, but this
will be done as an extra step in course of the entry code consolidation
work.
So the proper solution for this is to:
- Split kvm_async_pf_task_wait() into schedule and halt based waiting
interfaces which share the enqueueing code.
- Add comments (condensed form of this changelog) to spare others the
time waste and pain of reverse engineering all of this with the help of
uncomprehensible changelogs and code history.
- Invoke kvm_async_pf_task_wait_schedule() from kvm_handle_page_fault(),
user mode and schedulable kernel side async page faults (#1, #2a, #2b)
- Invoke kvm_async_pf_task_wait_halt() for the non schedulable kernel
case (#2c).
For this case also remove the rcu_irq_exit()/enter() pair around the
halt as it is just a pointless exercise:
- vCPUs can VMEXIT at any random point and can be scheduled out for
an arbitrary amount of time by the host and this is not any
different except that it voluntary triggers the exit via halt.
- The interrupted context could have RCU watching already. So the
rcu_irq_exit() before the halt is not gaining anything aside of
confusing the reader. Claiming that this might prevent RCU stalls
is just an illusion.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Alexandre Chartre <alexandre.chartre@oracle.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200505134059.262701431@linutronix.de
KVM overloads #PF to indicate two types of not-actually-page-fault
events. Right now, the KVM guest code intercepts them by modifying
the IDT and hooking the #PF vector. This makes the already fragile
fault code even harder to understand, and it also pollutes call
traces with async_page_fault and do_async_page_fault for normal page
faults.
Clean it up by moving the logic into do_page_fault() using a static
branch. This gets rid of the platform trap_init override mechanism
completely.
[ tglx: Fixed up 32bit, removed error code from the async functions and
massaged coding style ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Alexandre Chartre <alexandre.chartre@oracle.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200505134059.169270470@linutronix.de
Two new stats for exposing halt-polling cpu usage:
halt_poll_success_ns
halt_poll_fail_ns
Thus sum of these 2 stats is the total cpu time spent polling. "success"
means the VCPU polled until a virtual interrupt was delivered. "fail"
means the VCPU had to schedule out (either because the maximum poll time
was reached or it needed to yield the CPU).
To avoid touching every arch's kvm_vcpu_stat struct, only update and
export halt-polling cpu usage stats if we're on x86.
Exporting cpu usage as a u64 and in nanoseconds means we will overflow at
~500 years, which seems reasonably large.
Signed-off-by: David Matlack <dmatlack@google.com>
Signed-off-by: Jon Cargille <jcargill@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Message-Id: <20200508182240.68440-1-jcargill@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The hrtimer used to emulate the VMX-preemption timer must be pinned to
the same logical processor as the vCPU thread to be interrupted if we
want to have any hope of adhering to the architectural specification
of the VMX-preemption timer. Even with this change, the emulated
VMX-preemption timer VM-exit occasionally arrives too late.
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20200508203643.85477-4-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Prepare for migration of this hrtimer, by changing it from relative to
absolute. (I couldn't get migration to work with a relative timer.)
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20200508203643.85477-3-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The PINNED bit is ignored by hrtimer_init. It is only considered when
starting the timer.
When the hrtimer isn't pinned to the same logical processor as the
vCPU thread to be interrupted, the emulated VMX-preemption timer
often fails to adhere to the architectural specification.
Fixes: f15a75eedc ("KVM: nVMX: make emulated nested preemption timer pinned")
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20200508203643.85477-2-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Remove a 'struct kvm_x86_ops' param that got left behind when the nested
ops were moved to their own struct.
Fixes: 33b2217245 ("KVM: x86: move nested-related kvm_x86_ops to a separate struct")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200506204653.14683-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This has already been handled in the prior call to svm_clear_vintr().
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <1588771076-73790-5-git-send-email-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Code clean up and remove unnecessary intercept check for
INTERCEPT_VINTR.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <1588771076-73790-4-git-send-email-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch implements a fastpath for the preemption timer vmexit. The vmexit
can be handled quickly so it can be performed with interrupts off and going
back directly to the guest.
Testing on SKX Server.
cyclictest in guest(w/o mwait exposed, adaptive advance lapic timer is default -1):
5540.5ns -> 4602ns 17%
kvm-unit-test/vmexit.flat:
w/o avanced timer:
tscdeadline_immed: 3028.5 -> 2494.75 17.6%
tscdeadline: 5765.7 -> 5285 8.3%
w/ adaptive advance timer default -1:
tscdeadline_immed: 3123.75 -> 2583 17.3%
tscdeadline: 4663.75 -> 4537 2.7%
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1588055009-12677-8-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch implements a fast path for emulation of writes to the TSCDEADLINE
MSR. Besides shortcutting various housekeeping tasks in the vCPU loop,
the fast path can also deliver the timer interrupt directly without going
through KVM_REQ_PENDING_TIMER because it runs in vCPU context.
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1588055009-12677-7-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Replace the ad hoc test in vmx_set_hv_timer with a test in the caller,
start_hv_timer. This test is not Intel-specific and would be duplicated
when introducing the fast path for the TSC deadline MSR.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
While optimizing posted-interrupt delivery especially for the timer
fastpath scenario, I measured kvm_x86_ops.deliver_posted_interrupt()
to introduce substantial latency because the processor has to perform
all vmentry tasks, ack the posted interrupt notification vector,
read the posted-interrupt descriptor etc.
This is not only slow, it is also unnecessary when delivering an
interrupt to the current CPU (as is the case for the LAPIC timer) because
PIR->IRR and IRR->RVI synchronization is already performed on vmentry
Therefore skip kvm_vcpu_trigger_posted_interrupt in this case, and
instead do vmx_sync_pir_to_irr() on the EXIT_FASTPATH_REENTER_GUEST
fastpath as well.
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1588055009-12677-6-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Adds a fastpath_t typedef since enum lines are a bit long, and replace
EXIT_FASTPATH_SKIP_EMUL_INS with two new exit_fastpath_completion enum values.
- EXIT_FASTPATH_EXIT_HANDLED kvm will still go through it's full run loop,
but it would skip invoking the exit handler.
- EXIT_FASTPATH_REENTER_GUEST complete fastpath, guest can be re-entered
without invoking the exit handler or going
back to vcpu_run
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1588055009-12677-4-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce kvm_vcpu_exit_request() helper, we need to check some conditions
before enter guest again immediately, we skip invoking the exit handler and
go through full run loop if complete fastpath but there is stuff preventing
we enter guest again immediately.
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1588055009-12677-5-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use __print_flags() to display the names of VMX flags in VM-Exit traces
and strip the flags when printing the basic exit reason, e.g. so that a
failed VM-Entry due to invalid guest state gets recorded as
"INVALID_STATE FAILED_VMENTRY" instead of "0x80000021".
Opportunstically fix misaligned variables in the kvm_exit and
kvm_nested_vmexit_inject tracepoints.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200508235348.19427-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce generic fastpath handler to handle MSR fastpath, VMX-preemption
timer fastpath etc; move it after vmx_complete_interrupts() in order to
catch events delivered to the guest, and abort the fast path in later
patches. While at it, move the kvm_exit tracepoint so that it is printed
for fastpath vmexits as well.
There is no observed performance effect for the IPI fastpath after this patch.
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Suggested-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <1588055009-12677-2-git-send-email-wanpengli@tencent.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Don't propagate GUEST_SYSENTER_* from vmcs02 to vmcs12 on nested VM-Exit
as the vmcs12 fields are updated in vmx_set_msr(), and writes to the
corresponding MSRs are always intercepted by KVM when running L2.
Dropping the propagation was intended to be done in the same commit that
added vmcs12 writes in vmx_set_msr()[1], but for reasons unknown was
only shuffled around[2][3].
[1] https://patchwork.kernel.org/patch/10933215
[2] https://patchwork.kernel.org/patch/10933215/#22682289
[3] https://lore.kernel.org/patchwork/patch/1088643
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428231025.12766-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Explicitly truncate the data written to vmcs.SYSENTER_EIP/ESP on WRMSR
if the virtual CPU doesn't support 64-bit mode. The SYSENTER address
fields in the VMCS are natural width, i.e. bits 63:32 are dropped if the
CPU doesn't support Intel 64 architectures. This behavior is visible to
the guest after a VM-Exit/VM-Exit roundtrip, e.g. if the guest sets bits
63:32 in the actual MSR.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428231025.12766-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Improve handle_external_interrupt_irqoff inline assembly in several ways:
- remove unneeded %c operand modifiers and "$" prefixes
- use %rsp instead of _ASM_SP, since we are in CONFIG_X86_64 part
- use $-16 immediate to align %rsp
- remove unneeded use of __ASM_SIZE macro
- define "ss" named operand only for X86_64
The patch introduces no functional changes.
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Message-Id: <20200504155706.2516956-1-ubizjak@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The index returned by kvm_async_pf_gfn_slot() will be removed when an
async pf gfn is going to be removed. However kvm_async_pf_gfn_slot()
is not reliable in that it can return the last key it loops over even
if the gfn is not found in the async gfn array. It should never
happen, but it's still better to sanity check against that to make
sure no unexpected gfn will be removed.
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200416155910.267514-1-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Forcing the ASYNC_PF_PER_VCPU to be power of two is much easier to be
used rather than calling roundup_pow_of_two() from time to time. Do
this by adding a BUILD_BUG_ON() inside the hash function.
Another point is that generally async pf does not allow concurrency
over ASYNC_PF_PER_VCPU after all (see kvm_setup_async_pf()), so it
does not make much sense either to have it not a power of two or some
of the entries will definitely be wasted.
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200416155859.267366-1-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
POP [mem] defaults to the word size, and the only legal non-default
size is 16 bits, e.g. a 32-bit POP will #UD in 64-bit mode and vice
versa, no need to use __ASM_SIZE macro to force operating mode.
Changes since v1:
- Fix commit message.
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Message-Id: <20200427205035.1594232-1-ubizjak@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a helper, mmu_alloc_root(), to consolidate the allocation of a root
shadow page, which has the same basic mechanics for all flavors of TDP
and shadow paging.
Note, __pa(sp->spt) doesn't need to be protected by mmu_lock, sp->spt
points at a kernel page.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428023714.31923-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Replace KVM's PT_PAGE_TABLE_LEVEL, PT_DIRECTORY_LEVEL and PT_PDPE_LEVEL
with the kernel's PG_LEVEL_4K, PG_LEVEL_2M and PG_LEVEL_1G. KVM's
enums are borderline impossible to remember and result in code that is
visually difficult to audit, e.g.
if (!enable_ept)
ept_lpage_level = 0;
else if (cpu_has_vmx_ept_1g_page())
ept_lpage_level = PT_PDPE_LEVEL;
else if (cpu_has_vmx_ept_2m_page())
ept_lpage_level = PT_DIRECTORY_LEVEL;
else
ept_lpage_level = PT_PAGE_TABLE_LEVEL;
versus
if (!enable_ept)
ept_lpage_level = 0;
else if (cpu_has_vmx_ept_1g_page())
ept_lpage_level = PG_LEVEL_1G;
else if (cpu_has_vmx_ept_2m_page())
ept_lpage_level = PG_LEVEL_2M;
else
ept_lpage_level = PG_LEVEL_4K;
No functional change intended.
Suggested-by: Barret Rhoden <brho@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428005422.4235-4-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename PT_MAX_HUGEPAGE_LEVEL to KVM_MAX_HUGEPAGE_LEVEL and make it a
separate define in anticipation of dropping KVM's PT_*_LEVEL enums in
favor of the kernel's PG_LEVEL_* enums.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428005422.4235-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Change the PSE hugepage handling in walk_addr_generic() to fire on any
page level greater than PT_PAGE_TABLE_LEVEL, a.k.a. PG_LEVEL_4K. PSE
paging only has two levels, so "== 2" and "> 1" are functionally the
same, i.e. this is a nop.
A future patch will drop KVM's PT_*_LEVEL enums in favor of the kernel's
PG_LEVEL_* enums, at which point "walker->level == PG_LEVEL_2M" is
semantically incorrect (though still functionally ok).
No functional change intended.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428005422.4235-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
vcpu->arch.guest_xstate_size lost its only user since commit df1daba7d1
("KVM: x86: support XSAVES usage in the host"), so clean it up.
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-Id: <20200429154312.1411-1-xiaoyao.li@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use an enum for passing around the failure code for a failed VM-Enter
that results in VM-Exit to provide a level of indirection from the final
resting place of the failure code, vmcs.EXIT_QUALIFICATION. The exit
qualification field is an unsigned long, e.g. passing around
'u32 exit_qual' throws up red flags as it suggests KVM may be dropping
bits when reporting errors to L1. This is a red herring because the
only defined failure codes are 0, 2, 3, and 4, i.e. don't come remotely
close to overflowing a u32.
Setting vmcs.EXIT_QUALIFICATION on entry failure is further complicated
by the MSR load list, which returns the (1-based) entry that failed, and
the number of MSRs to load is a 32-bit VMCS field. At first blush, it
would appear that overflowing a u32 is possible, but the number of MSRs
that can be loaded is hardcapped at 4096 (limited by MSR_IA32_VMX_MISC).
In other words, there are two completely disparate types of data that
eventually get stuffed into vmcs.EXIT_QUALIFICATION, neither of which is
an 'unsigned long' in nature. This was presumably the reasoning for
switching to 'u32' when the related code was refactored in commit
ca0bde28f2 ("kvm: nVMX: Split VMCS checks from nested_vmx_run()").
Using an enum for the failure code addresses the technically-possible-
but-will-never-happen scenario where Intel defines a failure code that
doesn't fit in a 32-bit integer. The enum variables and values will
either be automatically sized (gcc 5.4 behavior) or be subjected to some
combination of truncation. The former case will simply work, while the
latter will trigger a compile-time warning unless the compiler is being
particularly unhelpful.
Separating the failure code from the failed MSR entry allows for
disassociating both from vmcs.EXIT_QUALIFICATION, which avoids the
conundrum where KVM has to choose between 'u32 exit_qual' and tracking
values as 'unsigned long' that have no business being tracked as such.
To cement the split, set vmcs12->exit_qualification directly from the
entry error code or failed MSR index instead of bouncing through a local
variable.
Opportunistically rename the variables in load_vmcs12_host_state() and
vmx_set_nested_state() to call out that they're ignored, set exit_reason
on demand on nested VM-Enter failure, and add a comment in
nested_vmx_load_msr() to call out that returning 'i + 1' can't wrap.
No functional change intended.
Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200511220529.11402-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Snapshot the TDP level now that it's invariant (SVM) or dependent only
on host capabilities and guest CPUID (VMX). This avoids having to call
kvm_x86_ops.get_tdp_level() when initializing a TDP MMU and/or
calculating the page role, and thus avoids the associated retpoline.
Drop the WARN in vmx_get_tdp_level() as updating CPUID while L2 is
active is legal, if dodgy.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-11-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Separate the "core" TDP level handling from the nested EPT path to make
it clear that kvm_x86_ops.get_tdp_level() is used if and only if nested
EPT is not in use (kvm_init_shadow_ept_mmu() calculates the level from
the passed in vmcs12->eptp). Add a WARN_ON() to enforce that the
kvm_x86_ops hook is not called for nested EPT.
This sets the stage for snapshotting the non-"nested EPT" TDP page level
during kvm_cpuid_update() to avoid the retpoline associated with
kvm_x86_ops.get_tdp_level() when resetting the MMU, a relatively
frequent operation when running a nested guest.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-10-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move CR0 caching into the standard register caching mechanism in order
to take advantage of the availability checks provided by regs_avail.
This avoids multiple VMREADs in the (uncommon) case where kvm_read_cr0()
is called multiple times in a single VM-Exit, and more importantly
eliminates a kvm_x86_ops hook, saves a retpoline on SVM when reading
CR0, and squashes the confusing naming discrepancy of "cache_reg" vs.
"decache_cr0_guest_bits".
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-8-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move CR4 caching into the standard register caching mechanism in order
to take advantage of the availability checks provided by regs_avail.
This avoids multiple VMREADs and retpolines (when configured) during
nested VMX transitions as kvm_read_cr4_bits() is invoked multiple times
on each transition, e.g. when stuffing CR0 and CR3.
As an added bonus, this eliminates a kvm_x86_ops hook, saves a retpoline
on SVM when reading CR4, and squashes the confusing naming discrepancy
of "cache_reg" vs. "decache_cr4_guest_bits".
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-7-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Unconditionally check the validity of the incoming CR3 during nested
VM-Enter/VM-Exit to avoid invoking kvm_read_cr3() in the common case
where the guest isn't using PAE paging. If vmcs.GUEST_CR3 hasn't yet
been cached (common case), kvm_read_cr3() will trigger a VMREAD. The
VMREAD (~30 cycles) alone is likely slower than nested_cr3_valid()
(~5 cycles if vcpu->arch.maxphyaddr gets a cache hit), and the poor
exchange only gets worse when retpolines are enabled as the call to
kvm_x86_ops.cache_reg() will incur a retpoline (60+ cycles).
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Save L1's TSC offset in 'struct kvm_vcpu_arch' and drop the kvm_x86_ops
hook read_l1_tsc_offset(). This avoids a retpoline (when configured)
when reading L1's effective TSC, which is done at least once on every
VM-Exit.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Skip the Indirect Branch Prediction Barrier that is triggered on a VMCS
switch when temporarily loading vmcs02 to synchronize it to vmcs12, i.e.
give copy_vmcs02_to_vmcs12_rare() the same treatment as
vmx_switch_vmcs().
Make vmx_vcpu_load() static now that it's only referenced within vmx.c.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200506235850.22600-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Skip the Indirect Branch Prediction Barrier that is triggered on a VMCS
switch when running with spectre_v2_user=on/auto if the switch is
between two VMCSes in the same guest, i.e. between vmcs01 and vmcs02.
The IBPB is intended to prevent one guest from attacking another, which
is unnecessary in the nested case as it's the same guest from KVM's
perspective.
This all but eliminates the overhead observed for nested VMX transitions
when running with CONFIG_RETPOLINE=y and spectre_v2_user=on/auto, which
can be significant, e.g. roughly 3x on current systems.
Reported-by: Alexander Graf <graf@amazon.com>
Cc: KarimAllah Raslan <karahmed@amazon.de>
Cc: stable@vger.kernel.org
Fixes: 15d4507152 ("KVM/x86: Add IBPB support")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200501163117.4655-1-sean.j.christopherson@intel.com>
[Invert direction of bool argument. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use vmx_get_intr_info() when grabbing the cached vmcs.INTR_INFO in
handle_exception_nmi() to ensure the cache isn't stale. Bypassing the
caching accessor doesn't cause any known issues as the cache is always
refreshed by handle_exception_nmi_irqoff(), but the whole point of
adding the proper caching mechanism was to avoid such dependencies.
Fixes: 8791585837 ("KVM: VMX: Cache vmcs.EXIT_INTR_INFO using arch avail_reg flags")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200427171837.22613-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM is not handling the case where EIP wraps around the 32-bit address
space (that is, outside long mode). This is needed both in vmx.c
and in emulate.c. SVM with NRIPS is okay, but it can still print
an error to dmesg due to integer overflow.
Reported-by: Nick Peterson <everdox@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The use of any sort of waitqueue (simple or regular) for
wait/waking vcpus has always been an overkill and semantically
wrong. Because this is per-vcpu (which is blocked) there is
only ever a single waiting vcpu, thus no need for any sort of
queue.
As such, make use of the rcuwait primitive, with the following
considerations:
- rcuwait already provides the proper barriers that serialize
concurrent waiter and waker.
- Task wakeup is done in rcu read critical region, with a
stable task pointer.
- Because there is no concurrency among waiters, we need
not worry about rcuwait_wait_event() calls corrupting
the wait->task. As a consequence, this saves the locking
done in swait when modifying the queue. This also applies
to per-vcore wait for powerpc kvm-hv.
The x86 tscdeadline_latency test mentioned in 8577370fb0
("KVM: Use simple waitqueue for vcpu->wq") shows that, on avg,
latency is reduced by around 15-20% with this change.
Cc: Paul Mackerras <paulus@ozlabs.org>
Cc: kvmarm@lists.cs.columbia.edu
Cc: linux-mips@vger.kernel.org
Reviewed-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Message-Id: <20200424054837.5138-6-dave@stgolabs.net>
[Avoid extra logic changes. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add an argument to interrupt_allowed and nmi_allowed, to checking if
interrupt injection is blocked. Use the hook to handle the case where
an interrupt arrives between check_nested_events() and the injection
logic. Drop the retry of check_nested_events() that hack-a-fixed the
same condition.
Blocking injection is also a bit of a hack, e.g. KVM should do exiting
and non-exiting interrupt processing in a single pass, but it's a more
precise hack. The old comment is also misleading, e.g. KVM_REQ_EVENT is
purely an optimization, setting it on every run loop (which KVM doesn't
do) should not affect functionality, only performance.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-13-sean.j.christopherson@intel.com>
[Extend to SVM, add SMI and NMI. Even though NMI and SMI cannot come
asynchronously right now, making the fix generic is easy and removes a
special case. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use vmx_get_rflags() instead of manually reading vmcs.GUEST_RFLAGS when
querying RFLAGS.IF so that multiple checks against interrupt blocking in
a single run loop only require a single VMREAD.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-14-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use vmx_interrupt_blocked() instead of bouncing through
vmx_interrupt_allowed() when handling edge cases in vmx_handle_exit().
The nested_run_pending check in vmx_interrupt_allowed() should never
evaluate true in the VM-Exit path.
Hoist the WARN in handle_invalid_guest_state() up to vmx_handle_exit()
to enforce the above assumption for the !enable_vnmi case, and to detect
any other potential bugs with nested VM-Enter.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-12-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
WARN if a pending exception is coincident with an injected exception
before calling check_nested_events() so that the WARN will fire even if
inject_pending_event() bails early because check_nested_events() detects
the conflict. Bailing early isn't problematic (quite the opposite), but
suppressing the WARN is undesirable as it could mask a bug elsewhere in
KVM.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-11-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Short circuit vmx_check_nested_events() if an unblocked IRQ/NMI/SMI is
pending and needs to be injected into L2, priority between coincident
events is not dependent on exiting behavior.
Fixes: b518ba9fa6 ("KVM: nSVM: implement check_nested_events for interrupts")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Report interrupts as allowed when the vCPU is in L2 and L2 is being run with
exit-on-interrupts enabled and EFLAGS.IF=1 (either on the host or on the guest
according to VINTR). Interrupts are always unblocked from L1's perspective
in this case.
While moving nested_exit_on_intr to svm.h, use INTERCEPT_INTR properly instead
of assuming it's zero (which it is of course).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Check for an unblocked SMI in vmx_check_nested_events() so that pending
SMIs are correctly prioritized over IRQs and NMIs when the latter events
will trigger VM-Exit. This also fixes an issue where an SMI that was
marked pending while processing a nested VM-Enter wouldn't trigger an
immediate exit, i.e. would be incorrectly delayed until L2 happened to
take a VM-Exit.
Fixes: 64d6067057 ("KVM: x86: stubs for SMM support")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-10-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Short circuit vmx_check_nested_events() if an unblocked IRQ/NMI is
pending and needs to be injected into L2, priority between coincident
events is not dependent on exiting behavior.
Fixes: b6b8a1451f ("KVM: nVMX: Rework interception of IRQs and NMIs")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-9-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
