...now that the sub-routine follows standard calling conventions.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to make it callable from C code.

Note that because KVM chooses to be ultra paranoid about guest register
values, all callee-save registers are still cleared after VM-Exit even
though the host's values are now reloaded from the stack.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to prepare for making the assembly sub-routine callable from C code.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to prepare for making the sub-routine callable from C code.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to prepare for making the sub-routine callable from C code.  That
means returning the result in RAX.  Since RAX will be used to return the
result, use it as the scratch register as well to make the code readable
and to document that the scratch register is more or less arbitrary.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...now that the name is no longer usurped by a defunct helper function.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...now that the code is no longer tagged with STACK_FRAME_NON_STANDARD.
Arguably, providing __vmx_vcpu_run() to break up vmx_vcpu_run() is
valuable on its own, but the previous split was purposely made as small
as possible to limit the effects STACK_FRAME_NON_STANDARD.  In other
words, the current split is now completely arbitrary and likely not the
most logical.

This also allows renaming ____vmx_vcpu_run() to __vmx_vcpu_run() in a
future patch.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

As evidenced by the myriad patches leading up to this moment, using
an inline asm blob for vCPU-run is nothing short of horrific.  It's also
been called "unholy", "an abomination" and likely a whole host of other
names that would violate the Code of Conduct if recorded here and now.

The code is relocated nearly verbatim, e.g. quotes, newlines, tabs and
__stringify need to be dropped, but other than those cosmetic changes
the only functional changees are to add the "call" and replace the final
"jmp" with a "ret".

Note that STACK_FRAME_NON_STANDARD is also dropped from __vmx_vcpu_run().
Suggested-by: Andi Kleen <ak@linux.intel.com>
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...in preparation for moving to a proper assembly sub-routnine.
vCPU-run isn't a leaf function since it calls vmx_update_host_rsp()
and vmx_vmenter().  And since we need to save/restore RBP anyways,
unconditionally creating the frame costs a single MOV, i.e. don't
bother keying off CONFIG_FRAME_POINTER or using FRAME_BEGIN, etc...
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to prepare for moving the inline asm to a proper asm sub-routine.
Eliminating the immediates allows a nearly verbatim move, e.g. quotes,
newlines, tabs and __stringify need to be dropped, but other than those
cosmetic changes the only function change will be to replace the final
"jmp" with a "ret".
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Declaring the VCPU_REGS_* as enums allows for more robust C code, but it
prevents using the values in assembly files.  Expliciting #define the
indices in an asm-friendly file to prepare for VMX moving its transition
code to a proper assembly file, but keep the enums for general usage.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...now that all other references to struct vcpu_vmx have been removed.

Note that 'vmx' still needs to be passed into the asm blob in _ASM_ARG1
as it is consumed by vmx_update_host_rsp().  And similar to that code,
use _ASM_ARG2 in the assembly code to prepare for moving to proper asm,
while explicitly referencing the exact registers in the clobber list for
clarity in the short term and to avoid additional precompiler games.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

A failed VM-Enter (obviously) didn't succeed, meaning the CPU never
executed an instrunction in guest mode and so can't have changed the
general purpose registers.

In addition to saving some instructions in the VM-Fail case, this also
provides a separate path entirely and thus an opportunity to propagate
the fail condition to vmx->fail via register without introducing undue
pain.  Using a register, as opposed to directly referencing vmx->fail,
eliminates the need to pass the offset of 'fail', which will simplify
moving the code to proper assembly in future patches.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Switching the ordering allows for an out-of-line path for VM-Fail
that elides saving guest state but still shares the register clearing
with the VM-Exit path.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...and remove struct vcpu_vmx's temporary __launched variable.

Eliminating __launched is a bonus, the real motivation is to get to the
point where the only reference to struct vcpu_vmx in the asm code is
to vcpu.arch.regs, which will simplify moving the blob to a proper asm
file.  Note that also means this approach is deliberately different than
what is used in nested_vmx_check_vmentry_hw().

Use BL as it is a callee-save register in both 32-bit and 64-bit ABIs,
i.e. it can't be modified by vmx_update_host_rsp(), to avoid having to
temporarily save/restore the launched flag.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Providing a helper function to update HOST_RSP is visibly easier to
read, and more importantly (for the future) eliminates two arguments to
the VM-Enter assembly blob.  Reducing the number of arguments to the asm
blob is for all intents and purposes a prerequisite to moving the code
to a proper assembly routine.  It's not truly mandatory, but it greatly
simplifies the future code, and the cost of the extra CALL+RET is
negligible in the grand scheme.

Note that although _ASM_ARG[1-3] can be used in the inline asm itself,
the intput/output constraints need to be manually defined.  gcc will
actually compile with _ASM_ARG[1-3] specified as constraints, but what
it actually ends up doing with the bogus constraint is unknown.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to eliminate its parameter and struct vcpu_vmx offset definition
from the assembly blob.  Accessing CR2 from C versus assembly doesn't
change the likelihood of taking a page fault (and modifying CR2) while
it's loaded with the guest's value, so long as we don't do anything
silly between accessing CR2 and VM-Enter/VM-Exit.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Currently, host_rsp is cached on a per-vCPU basis, i.e. it's stored in
struct vcpu_vmx.  In non-nested usage the caching is for all intents
and purposes 100% effective, e.g. only the first VMLAUNCH needs to
synchronize VMCS.HOST_RSP since the call stack to vmx_vcpu_run() is
identical each and every time.  But when running a nested guest, KVM
must invalidate the cache when switching the current VMCS as it can't
guarantee the new VMCS has the same HOST_RSP as the previous VMCS.  In
other words, the cache loses almost all of its efficacy when running a
nested VM.

Move host_rsp to struct vmcs_host_state, which is per-VMCS, so that it
is cached on a per-VMCS basis and restores its 100% hit rate when
nested VMs are in play.

Note that the host_rsp cache for vmcs02 essentially "breaks" when
nested early checks are enabled as nested_vmx_check_vmentry_hw() will
see a different RSP at the time of its VM-Enter.  While it's possible
to avoid even that VMCS.HOST_RSP synchronization, e.g. by employing a
dedicated VM-Exit stack, there is little motivation for doing so as
the overhead of two VMWRITEs (~55 cycles) is dwarfed by the overhead
of the extra VMX transition (600+ cycles) and is a proverbial drop in
the ocean relative to the total cost of a nested transtion (10s of
thousands of cycles).
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...and provide an explicit name for the constraint.  Naming the input
constraint makes the code self-documenting and also avoids the fragility
of numerically referring to constraints, e.g. %4 breaks badly whenever
the constraints are modified.

Explicitly using RDX was inherited from vCPU-run, i.e. completely
arbitrary.  Even vCPU-run doesn't truly need to explicitly use RDX, but
doing so is more robust as vCPU-run needs tight control over its
register usage.

Note that while the naming "conflict" between host_rsp and HOST_RSP
is slightly confusing, the former will be renamed slightly in a
future patch, at which point HOST_RSP is absolutely what is desired.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Temporarily propagating vmx->loaded_vmcs->launched to vmx->__launched
is not functionally necessary, but rather was done historically to
avoid passing both 'vmx' and 'loaded_vmcs' to the vCPU-run asm blob.
Nested early checks inherited this behavior by virtue of copy+paste.

A future patch will move HOST_RSP caching to be per-VMCS, i.e. store
'host_rsp' in loaded VMCS.  Now that the reference to 'vmx->fail' is
also gone from nested early checks, referencing 'loaded_vmcs' directly
means we can drop the 'vmx' reference when introducing per-VMCS RSP
caching.  And it means __launched can be dropped from struct vcpu_vmx
if/when vCPU-run receives similar treatment.

Note the use of a named register constraint for 'loaded_vmcs'.  Using
RCX to hold 'vmx' was inherited from vCPU-run.  In the vCPU-run case,
the scratch register needs to be explicitly defined as it is crushed
when loading guest state, i.e. deferring to the compiler would corrupt
the pointer.  Since nested early checks never loads guests state, it's
a-ok to let the compiler pick any register.  Naming the constraint
avoids the fragility of referencing constraints via %1, %2, etc.., which
breaks horribly when modifying constraints, and generally makes the asm
blob more readable.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...to take advantage of __GCC_ASM_FLAG_OUTPUTS__ when possible.
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Unlike the primary vCPU-run flow, the nested early checks code doesn't
actually want to propagate VM-Fail back to 'vmx'.  Yay copy+paste.

In additional to eliminating the need to clear vmx->fail before
returning, using a local boolean also drops a reference to 'vmx' in the
asm blob.  Dropping the reference to 'vmx' will save a register in the
long run as future patches will shift all pointer references from 'vmx'
to 'vmx->loaded_vmcs'.

Fixes: 52017608 ("KVM: nVMX: add option to perform early consistency checks via H/W")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Using %1 to reference RCX, i.e. the 'vmx' pointer', is obtuse and
fragile, e.g. it results in cryptic and infurating compile errors if the
output constraints are touched by anything more than a gentle breeze.
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...as it doesn't technically actually do anything non-standard with the
stack even though it modifies RSP in a weird way.  E.g. RSP is loaded
with VMCS.HOST_RSP if the VM-Enter gets far enough to trigger VM-Exit,
but it's simply reloaded with the current value.
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

RAX is not touched by nested_vmx_check_vmentry_hw(), directly or
indirectly (e.g. vmx_vmenter()).  Remove it from the clobber list.

Fixes: 52017608 ("KVM: nVMX: add option to perform early consistency checks via H/W")
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Per commit c2036300 ("KVM: VMX: Let gcc to choose which registers
to save (x86_64)"), the only reason RDX is saved/loaded to/from the
stack is because it was specified as an input, i.e. couldn't be marked
as clobbered (ignoring the fact that "saving" it to a dummy output
would indirectly mark it as clobbered).

Now that RDX is no longer an input, clobber it.
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Load RDX with the VMCS.HOST_RSP field encoding on-demand instead of
delegating to the compiler via an input constraint.  In addition to
saving one whole MOV instruction, this allows RDX to be properly
clobbered (in a future patch) instead of being saved/loaded to/from
the stack.

Despite nested_vmx_check_vmentry_hw() having similar code, leave it
alone, for now.  In that case, RDX is unconditionally used and isn't
clobbered, i.e. sending in HOST_RSP as an input is simpler.

Note that because HOST_RSP is an enum and not a define, it must be
redefined as an immediate instead of using __stringify(HOST_RSP).  The
naming "conflict" between host_rsp and HOST_RSP is slightly confusing,
but the former will be removed in a future patch, at which point
HOST_RSP is absolutely what is desired.
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

RSI is clobbered by the vCPU-run asm blob, but it's not marked as such,
probably because GCC doesn't let you mark inputs as clobbered.  "Save"
RSI to a dummy output so that GCC recognizes it as being clobbered.

Fixes: 773e8a04 ("x86/kvm: use Enlightened VMCS when running on Hyper-V")
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

In the vCPU-run asm blob, the guest's RCX is temporarily saved onto the
stack after VM-Exit as the exit flow must first load a register with a
pointer to the vCPU's save area in order to save the guest's registers.
RCX is arbitrarily designated as the scratch register.

Since the stack usage is to (1)save host, (2)save guest, (3)load host
and (4)load guest, the code can't conform to the stack's natural FIFO
semantics, i.e. it can't simply do PUSH/POP.  Regardless of whether it
is done for the host's value or guest's value, at some point the code
needs to access the stack using a non-traditional method, e.g. MOV
instead of POP.  vCPU-run opts to create a placeholder on the stack for
guest's RCX (by adjusting RSP) and saves RCX to its place immediately
after VM-Exit (via MOV).

In other words, the purpose of the first 'PUSH RCX' at the start of
the vCPU-run asm blob  is to adjust RSP down, i.e. there's no need to
actually access memory.  Use 'SUB $wordsize, RSP' instead of 'PUSH RCX'
to make it more obvious that the intent is simply to create a gap on
the stack for the guest's RCX.
Reviewed-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

...except RSP, which is restored by hardware as part of VM-Exit.

Paolo theorized that restoring registers from the stack after a VM-Exit
in lieu of zeroing them could lead to speculative execution with the
guest's values, e.g. if the stack accesses miss the L1 cache[1].
Zeroing XORs are dirt cheap, so just be ultra-paranoid.

Note that the scratch register (currently RCX) used to save/restore the
guest state is also zeroed as its host-defined value is loaded via the
stack, just with a MOV instead of a POP.

[1] https://patchwork.kernel.org/patch/10771539/#22441255

Fixes: 0cb5b306 ("kvm: vmx: Scrub hardware GPRs at VM-exit")
Cc: <stable@vger.kernel.org>
Cc: Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Nested early checks does a manual comparison of a VMCS' launched status
in its asm blob to execute the correct VM-Enter instruction, i.e.
VMLAUNCH vs. VMRESUME.  The launched flag is a bool, which is a typedef
of _Bool.  C99 does not define an exact size for _Bool, stating only
that is must be large enough to hold '0' and '1'.  Most, if not all,
compilers use a single byte for _Bool, including gcc[1].

The use of 'cmpl' instead of 'cmpb' was not deliberate, but rather the
result of a copy-paste as the asm blob was directly derived from the asm
blob for vCPU-run.

This has not caused any known problems, likely due to compilers aligning
variables to 4-byte or 8-byte boundaries and KVM zeroing out struct
vcpu_vmx during allocation.  I.e. vCPU-run accesses "junk" data, it just
happens to always be zero and so doesn't affect the result.

[1] https://gcc.gnu.org/ml/gcc-patches/2000-10/msg01127.html

Fixes: 52017608 ("KVM: nVMX: add option to perform early consistency checks via H/W")
Cc: <stable@vger.kernel.org>
Reviewed-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

The vCPU-run asm blob does a manual comparison of a VMCS' launched
status to execute the correct VM-Enter instruction, i.e. VMLAUNCH vs.
VMRESUME.  The launched flag is a bool, which is a typedef of _Bool.
C99 does not define an exact size for _Bool, stating only that is must
be large enough to hold '0' and '1'.  Most, if not all, compilers use
a single byte for _Bool, including gcc[1].

Originally, 'launched' was of type 'int' and so the asm blob used 'cmpl'
to check the launch status.  When 'launched' was moved to be stored on a
per-VMCS basis, struct vcpu_vmx's "temporary" __launched flag was added
in order to avoid having to pass the current VMCS into the asm blob.
The new  '__launched' was defined as a 'bool' and not an 'int', but the
'cmp' instruction was not updated.

This has not caused any known problems, likely due to compilers aligning
variables to 4-byte or 8-byte boundaries and KVM zeroing out struct
vcpu_vmx during allocation.  I.e. vCPU-run accesses "junk" data, it just
happens to always be zero and so doesn't affect the result.

[1] https://gcc.gnu.org/ml/gcc-patches/2000-10/msg01127.html

Fixes: d462b819 ("KVM: VMX: Keep list of loaded VMCSs, instead of vcpus")
Cc: <stable@vger.kernel.org>
Reviewed-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

This adds a test for the previous bug.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Pull dmaengine fixes from Vinod Koul:
 - Fix in at_xdmac fr wrongful channel state
 - Fix for imx driver for wrong callback invocation
 - Fix to bcm driver for interrupt race & transaction abort.
 - Fix in dmatest to abort in mapping error

* tag 'dmaengine-fix-5.0-rc6' of git://git.infradead.org/users/vkoul/slave-dma:
  dmaengine: dmatest: Abort test in case of mapping error
  dmaengine: bcm2835: Fix abort of transactions
  dmaengine: bcm2835: Fix interrupt race on RT
  dmaengine: imx-dma: fix wrong callback invoke
  dmaengine: at_xdmac: Fix wrongfull report of a channel as in use

Pull x86 fixes from Ingo Molnar:
 "A handful of fixes:

   - Fix an MCE corner case bug/crash found via MCE injection testing

   - Fix 5-level paging boot crash

   - Fix MCE recovery cache invalidation bug

   - Fix regression on Xen guests caused by a recent PMD level mremap
     speedup optimization"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mm: Make set_pmd_at() paravirt aware
  x86/mm/cpa: Fix set_mce_nospec()
  x86/boot/compressed/64: Do not corrupt EDX on EFER.LME=1 setting
  x86/MCE: Initialize mce.bank in the case of a fatal error in mce_no_way_out()

Pull irq fixes from Ingo Molnar:
 "irqchip driver fixes: most of them are race fixes for ARM GIC (General
  Interrupt Controller) variants, but also a fix for the ARM MMP
  (Marvell PXA168 et al) irqchip affecting OLPC keyboards"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/gic-v3-its: Fix ITT_entry_size accessor
  irqchip/mmp: Only touch the PJ4 IRQ & FIQ bits on enable/disable
  irqchip/gic-v3-its: Gracefully fail on LPI exhaustion
  irqchip/gic-v3-its: Plug allocation race for devices sharing a DevID
  irqchip/gic-v4: Fix occasional VLPI drop

Pull perf fixes from Ingo Molnar:
 "A couple of kernel side fixes:

   - Fix the Intel uncore driver on certain hardware configurations

   - Fix a CPU hotplug related memory allocation bug

   - Remove a spurious WARN()

  ... plus also a handful of perf tooling fixes"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf script python: Add Python3 support to tests/attr.py
  perf trace: Support multiple "vfs_getname" probes
  perf symbols: Filter out hidden symbols from labels
  perf symbols: Add fallback definitions for GELF_ST_VISIBILITY()
  tools headers uapi: Sync linux/in.h copy from the kernel sources
  perf clang: Do not use 'return std::move(something)'
  perf mem/c2c: Fix perf_mem_events to support powerpc
  perf tests evsel-tp-sched: Fix bitwise operator
  perf/core: Don't WARN() for impossible ring-buffer sizes
  perf/x86/intel: Delay memory deallocation until x86_pmu_dead_cpu()
  perf/x86/intel/uncore: Add Node ID mask

Pull locking fixes from Ingo Molnar:
 "An rtmutex (PI-futex) deadlock scenario fix, plus a locking
  documentation fix"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  futex: Handle early deadlock return correctly
  futex: Fix barrier comment

set_pmd_at() calls native_set_pmd() unconditionally on x86. This was
fine as long as only huge page entries were written via set_pmd_at(),
as Xen pv guests don't support those.

Commit 2c91bd4a ("mm: speed up mremap by 20x on large regions")
introduced a usage of set_pmd_at() possible on pv guests, leading to
failures like:

BUG: unable to handle kernel paging request at ffff888023e26778
#PF error: [PROT] [WRITE]
RIP: e030:move_page_tables+0x7c1/0xae0
move_vma.isra.3+0xd1/0x2d0
__se_sys_mremap+0x3c6/0x5b0
 do_syscall_64+0x49/0x100
entry_SYSCALL_64_after_hwframe+0x44/0xa9

Make set_pmd_at() paravirt aware by just letting it use set_pmd().

Fixes: 2c91bd4a ("mm: speed up mremap by 20x on large regions")
Reported-by: Sander Eikelenboom <linux@eikelenboom.it>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: xen-devel@lists.xenproject.org
Cc: boris.ostrovsky@oracle.com
Cc: sstabellini@kernel.org
Cc: hpa@zytor.com
Cc: bp@alien8.de
Cc: torvalds@linux-foundation.org
Link: https://lkml.kernel.org/r/20190210074056.11842-1-jgross@suse.com