GitHub/moto-9609/android_kernel_motorola_exynos9610.git
8 years agoMerge branch 'kvm-ppc-next' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus...
Paolo Bonzini [Mon, 11 Jul 2016 16:10:06 +0000 (18:10 +0200)]
Merge branch 'kvm-ppc-next' of git://git./linux/kernel/git/paulus/powerpc into HEAD

8 years agoKVM: VMX: introduce vm_{entry,exit}_control_reset_shadow
Paolo Bonzini [Thu, 7 Jul 2016 12:58:33 +0000 (14:58 +0200)]
KVM: VMX: introduce vm_{entry,exit}_control_reset_shadow

There is no reason to read the entry/exit control fields of the
VMCS and immediately write back the same value.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: nVMX: keep preemption timer enabled during L2 execution
Paolo Bonzini [Wed, 6 Jul 2016 11:23:51 +0000 (13:23 +0200)]
KVM: nVMX: keep preemption timer enabled during L2 execution

Because the vmcs12 preemption timer is emulated through a separate hrtimer,
we can keep on using the preemption timer in the vmcs02 to emulare L1's
TSC deadline timer.

However, the corresponding bit in the pin-based execution control field
must be kept consistent between vmcs01 and vmcs02.  On vmentry we copy
it into the vmcs02; on vmexit the preemption timer must be disabled in
the vmcs01 if a preemption timer vmexit happened while in guest mode.

The preemption timer value in the vmcs02 is set by vmx_vcpu_run, so it
need not be considered in prepare_vmcs02.

Cc: Yunhong Jiang <yunhong.jiang@intel.com>
Cc: Haozhong Zhang <haozhong.zhang@intel.com>
Tested-by: Wanpeng Li <kernellwp@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: nVMX: avoid incorrect preemption timer vmexit in nested guest
Wanpeng Li [Wed, 6 Jul 2016 10:29:58 +0000 (18:29 +0800)]
KVM: nVMX: avoid incorrect preemption timer vmexit in nested guest

The preemption timer for nested VMX is emulated by hrtimer which is started on L2
entry, stopped on L2 exit and evaluated via the check_nested_events hook. However,
nested_vmx_exit_handled is always returning true for preemption timer vmexit.  Then,
the L1 preemption timer vmexit is captured and be treated as a L2 preemption
timer vmexit, causing NULL pointer dereferences or worse in the L1 guest's
vmexit handler:

    BUG: unable to handle kernel NULL pointer dereference at           (null)
    IP: [<          (null)>]           (null)
    PGD 0
    Oops: 0010 [#1] SMP
    Call Trace:
     ? kvm_lapic_expired_hv_timer+0x47/0x90 [kvm]
     handle_preemption_timer+0xe/0x20 [kvm_intel]
     vmx_handle_exit+0x169/0x15a0 [kvm_intel]
     ? kvm_arch_vcpu_ioctl_run+0xd5d/0x19d0 [kvm]
     kvm_arch_vcpu_ioctl_run+0xdee/0x19d0 [kvm]
     ? kvm_arch_vcpu_ioctl_run+0xd5d/0x19d0 [kvm]
     ? vcpu_load+0x1c/0x60 [kvm]
     ? kvm_arch_vcpu_load+0x57/0x260 [kvm]
     kvm_vcpu_ioctl+0x2d3/0x7c0 [kvm]
     do_vfs_ioctl+0x96/0x6a0
     ? __fget_light+0x2a/0x90
     SyS_ioctl+0x79/0x90
     do_syscall_64+0x68/0x180
     entry_SYSCALL64_slow_path+0x25/0x25
    Code:  Bad RIP value.
    RIP  [<          (null)>]           (null)
     RSP <ffff8800b5263c48>
    CR2: 0000000000000000
    ---[ end trace 9c70c48b1a2bc66e ]---

This can be reproduced readily by preemption timer enabled on L0 and disabled
on L1.

Return false since preemption timer vmexits must never be reflected to L2.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Yunhong Jiang <yunhong.jiang@intel.com>
Cc: Jan Kiszka <jan.kiszka@siemens.com>
Cc: Haozhong Zhang <haozhong.zhang@intel.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: VMX: reflect broken preemption timer in vmcs_config
Paolo Bonzini [Fri, 8 Jul 2016 09:53:38 +0000 (11:53 +0200)]
KVM: VMX: reflect broken preemption timer in vmcs_config

Simplify cpu_has_vmx_preemption_timer.  This is consistent with the
rest of setup_vmcs_config and preparatory for the next patch.

Tested-by: Wanpeng Li <kernellwp@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Emulate generic QEMU machine on r6 T&E
James Hogan [Mon, 4 Jul 2016 18:35:15 +0000 (19:35 +0100)]
MIPS: KVM: Emulate generic QEMU machine on r6 T&E

Default the guest PRId register to represent a generic QEMU machine
instead of a 24kc on MIPSr6. 24kc isn't supported by r6 Linux kernels.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Decode RDHWR more strictly
James Hogan [Mon, 4 Jul 2016 18:35:14 +0000 (19:35 +0100)]
MIPS: KVM: Decode RDHWR more strictly

When KVM emulates the RDHWR instruction, decode the instruction more
strictly. The rs field (bits 25:21) should be zero, as should bits 10:9.
Bits 8:6 is the register select field in MIPSr6, so we aren't strict
about those bits (no other operations should use that encoding space).

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Recognise r6 CACHE encoding
James Hogan [Mon, 4 Jul 2016 18:35:13 +0000 (19:35 +0100)]
MIPS: KVM: Recognise r6 CACHE encoding

Recognise the new MIPSr6 CACHE instruction encoding rather than the
pre-r6 one when an r6 kernel is being built. A SPECIAL3 opcode is used
and the immediate field is reduced to 9 bits wide since MIPSr6.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Support r6 compact branch emulation
James Hogan [Mon, 4 Jul 2016 18:35:12 +0000 (19:35 +0100)]
MIPS: KVM: Support r6 compact branch emulation

Add support in KVM for emulation of instructions in the forbidden slot
of MIPSr6 compact branches. If we hit an exception on the forbidden
slot, then the branch must not have been taken, which makes calculation
of the resume PC trivial.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Don't save/restore lo/hi for r6
James Hogan [Mon, 4 Jul 2016 18:35:11 +0000 (19:35 +0100)]
MIPS: KVM: Don't save/restore lo/hi for r6

MIPSr6 doesn't have lo/hi registers, so don't bother saving or
restoring them, and don't expose them to userland with the KVM ioctl
interface either.

In fact the lo/hi registers aren't callee saved in the MIPS ABIs anyway,
so there is no need to preserve the host lo/hi values at all when
transitioning to and from the guest (which happens via a function call).

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Fix pre-r6 ll/sc instructions on r6
James Hogan [Mon, 4 Jul 2016 18:35:10 +0000 (19:35 +0100)]
MIPS: KVM: Fix pre-r6 ll/sc instructions on r6

The atomic KVM register access macros in kvm_host.h (for the guest Cause
register with KVM in trap & emulate mode) use ll/sc instructions,
however they still .set mips3, which causes pre-MIPSr6 instruction
encodings to be emitted, even for a MIPSr6 build.

Fix it to use MIPS_ISA_ARCH_LEVEL as other parts of arch/mips already
do.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Fix fpu.S misassembly with r6
James Hogan [Mon, 4 Jul 2016 18:35:09 +0000 (19:35 +0100)]
MIPS: KVM: Fix fpu.S misassembly with r6

__kvm_save_fpu and __kvm_restore_fpu use .set mips64r2 so that they can
access the odd FPU registers as well as the even, however this causes
misassembly of the return instruction on MIPSr6.

Fix by replacing .set mips64r2 with .set fp=64, which doesn't change the
architecture revision.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: inst.h: Rename cbcond{0,1}_op to pop{1,3}0_op
Paul Burton [Mon, 4 Jul 2016 18:35:08 +0000 (19:35 +0100)]
MIPS: inst.h: Rename cbcond{0,1}_op to pop{1,3}0_op

The opcodes currently defined in inst.h as cbcond0_op & cbcond1_op are
actually defined in the MIPS base instruction set manuals as pop10 &
pop30 respectively. Rename them as such, for consistency with the
documentation.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: inst.h: Rename b{eq,ne}zcji[al]c_op to pop{6,7}6_op
Paul Burton [Mon, 4 Jul 2016 18:35:07 +0000 (19:35 +0100)]
MIPS: inst.h: Rename b{eq,ne}zcji[al]c_op to pop{6,7}6_op

The opcodes currently defined in inst.h as beqzcjic_op & bnezcjialc_op
are actually defined in the MIPS base instruction set manuals as pop66 &
pop76 respectively. Rename them as such, for consistency with the
documentation.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Save k0 straight into VCPU structure
James Hogan [Thu, 23 Jun 2016 16:34:47 +0000 (17:34 +0100)]
MIPS: KVM: Save k0 straight into VCPU structure

Currently on a guest exception the guest's k0 register is saved to the
scratch temp register and the guest k1 saved to the exception base
address + 0x3000 using k0 to extract the Exception Base field of the
EBase register and as the base operand to the store. Both are then
copied into the VCPU structure after the other general purpose registers
have been saved there.

This bouncing to exception base + 0x3000 is not actually necessary as
the VCPU pointer can be determined and written through just as easily
with only a single spare register. The VCPU pointer is already needed in
k1 for saving the other GP registers, so lets save the guest k0 register
straight into the VCPU structure through k1, first saving k1 into the
scratch temp register instead of k0.

This could potentially pave the way for having a single exception base
area for use by all guests.

The ehb after saving the k register to the scratch temp register is also
delayed until just before it needs to be read back.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Relative branch to common exit handler
James Hogan [Thu, 23 Jun 2016 16:34:46 +0000 (17:34 +0100)]
MIPS: KVM: Relative branch to common exit handler

Use a relative branch to get from the individual exception vectors to
the common guest exit handler, rather than loading the address of the
exit handler and jumping to it.

This is made easier due to the fact we are now generating the entry code
dynamically. This will also allow the exception code to be further
reduced in future patches.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Dynamically choose scratch registers
James Hogan [Thu, 23 Jun 2016 16:34:45 +0000 (17:34 +0100)]
MIPS: KVM: Dynamically choose scratch registers

Scratch cop0 registers are needed by KVM to be able to save/restore all
the GPRs, including k0/k1, and for storing the VCPU pointer. However no
registers are universally suitable for these purposes, so the decision
should be made at runtime.

Until now, we've used DDATA_LO to store the VCPU pointer, and ErrorEPC
as a temporary. It could be argued that this is abuse of those
registers, and DDATA_LO is known not to be usable on certain
implementations (Cavium Octeon). If KScratch registers are present, use
them instead.

We save & restore the temporary register in addition to the VCPU pointer
register when using a KScratch register for it, as it may be used for
normal host TLB handling too.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Drop redundant restore of DDATA_LO
James Hogan [Thu, 23 Jun 2016 16:34:44 +0000 (17:34 +0100)]
MIPS: KVM: Drop redundant restore of DDATA_LO

On return from the exit handler to the host (without re-entering the
guest) we restore the saved value of the DDATA_LO register which we use
as a scratch register. However we've already restored it ready for
calling the exit handler so there is no need to do it again, so drop
that code.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Check MSA presence at uasm time
James Hogan [Thu, 23 Jun 2016 16:34:43 +0000 (17:34 +0100)]
MIPS: KVM: Check MSA presence at uasm time

Check for presence of MSA at uasm assembly time rather than at runtime
in the generated KVM host entry code. This optimises the guest exit path
by eliminating the MSA code entirely if not present, and eliminating the
read of Config3.MSAP and conditional branch if MSA is present.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Omit FPU handling entry code if possible
James Hogan [Thu, 23 Jun 2016 16:34:42 +0000 (17:34 +0100)]
MIPS: KVM: Omit FPU handling entry code if possible

The FPU handling code on entry from guest is unnecessary if no FPU is
present, so allow it to be dropped at uasm assembly time.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Drop now unused asm offsets
James Hogan [Thu, 23 Jun 2016 16:34:41 +0000 (17:34 +0100)]
MIPS: KVM: Drop now unused asm offsets

Now that locore.S is converted to uasm, remove a bunch of the assembly
offset definitions created by asm-offsets.c, including the CPUINFO_ ones
for reading the variable asid mask, and the non FPU/MSA related VCPU_
definitions. KVM's fpu.S and msa.S still use the remaining definitions.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Add dumping of generated entry code
James Hogan [Thu, 23 Jun 2016 16:34:40 +0000 (17:34 +0100)]
MIPS: KVM: Add dumping of generated entry code

Dump the generated entry code with pr_debug(), similar to how it is done
in tlbex.c, so it can be more easily debugged.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS; KVM: Convert exception entry to uasm
James Hogan [Thu, 23 Jun 2016 16:34:39 +0000 (17:34 +0100)]
MIPS; KVM: Convert exception entry to uasm

Convert the whole of locore.S (assembly to enter guest and handle
exception entry) to be generated dynamically with uasm. This is done
with minimal changes to the resulting code.

The main changes are:
- Some constants are generated by uasm using LUI+ADDIU instead of
  LUI+ORI.
- Loading of lo and hi are swapped around in vcpu_run but not when
  resuming the guest after an exit. Both bits of logic are now generated
  by the same code.
- Register MOVEs in uasm use different ADDU operand ordering to GNU as,
  putting zero register into rs instead of rt.
- The JALR.HB to call the C exit handler is switched to JALR, since the
  hazard barrier would appear to be unnecessary.

This will allow further optimisation in the future to dynamically handle
the capabilities of the CPU.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: uasm: Add r6 MUL encoding
James Hogan [Thu, 23 Jun 2016 16:34:38 +0000 (17:34 +0100)]
MIPS: uasm: Add r6 MUL encoding

Add the R6 MUL instruction encoding for 3 operand signed multiply to
uasm so that KVM can use uasm for generating its entry point code at
runtime on R6.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: uasm: Add MTHI/MTLO instructions
James Hogan [Thu, 23 Jun 2016 16:34:37 +0000 (17:34 +0100)]
MIPS: uasm: Add MTHI/MTLO instructions

Add MTHI/MTLO instructions for writing to the hi & lo registers to uasm
so that KVM can use uasm for generating its entry point code at runtime.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: uasm: Add DI instruction
James Hogan [Thu, 23 Jun 2016 16:34:36 +0000 (17:34 +0100)]
MIPS: uasm: Add DI instruction

Add DI instruction for disabling interrupts to uasm so that KVM can use
uasm for generating its entry point code at runtime.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: uasm: Add CFCMSA/CTCMSA instructions
James Hogan [Thu, 23 Jun 2016 16:34:35 +0000 (17:34 +0100)]
MIPS: uasm: Add CFCMSA/CTCMSA instructions

Add CFCMSA/CTCMSA instructions for accessing MSA control registers to
uasm so that KVM can use uasm for generating its entry point code at
runtime.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: uasm: Add CFC1/CTC1 instructions
James Hogan [Thu, 23 Jun 2016 16:34:34 +0000 (17:34 +0100)]
MIPS: uasm: Add CFC1/CTC1 instructions

Add CFC1/CTC1 instructions for accessing FP control registers to uasm so
that KVM can use uasm for generating its entry point code at runtime.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMerge tag 'kvm-s390-next-4.8-3' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 5 Jul 2016 12:42:00 +0000 (14:42 +0200)]
Merge tag 'kvm-s390-next-4.8-3' of git://git./linux/kernel/git/kvms390/linux into HEAD

KVM: s390: Fix for kvm/next (4.8) part 3

This contains a fix for PER ifetch events. As we now have a handler
for a problem state instruction (sthyi) that could be stepped with a
debugger we should try to do the right thing regarding PER in our
instruction handlers.  With this fix the handling for intercepted
instructions is fixed in general, thus fixing other oddball cases as
well (e.g. kprobes single stepping)

8 years agoKVM: x86: Use ARRAY_SIZE instead of dividing sizeof array with sizeof an element
Wei Yongjun [Mon, 4 Jul 2016 15:13:07 +0000 (15:13 +0000)]
KVM: x86: Use ARRAY_SIZE instead of dividing sizeof array with sizeof an element

Use ARRAY_SIZE instead of dividing sizeof array with sizeof an element

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: MMU: try to fix up page faults before giving up
Paolo Bonzini [Tue, 7 Jun 2016 15:51:18 +0000 (17:51 +0200)]
KVM: MMU: try to fix up page faults before giving up

The vGPU folks would like to trap the first access to a BAR by setting
vm_ops on the VMAs produced by mmap-ing a VFIO device.  The fault handler
then can use remap_pfn_range to place some non-reserved pages in the VMA.

This kind of VM_PFNMAP mapping is not handled by KVM, but follow_pfn
and fixup_user_fault together help supporting it.  The patch also supports
VM_MIXEDMAP vmas where the pfns are not reserved and thus subject to
reference counting.

Cc: Xiao Guangrong <guangrong.xiao@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Tested-by: Neo Jia <cjia@nvidia.com>
Reported-by: Kirti Wankhede <kwankhede@nvidia.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: MMU: prepare to support mapping of VM_IO and VM_PFNMAP frames
Paolo Bonzini [Tue, 7 Jun 2016 14:22:47 +0000 (16:22 +0200)]
KVM: MMU: prepare to support mapping of VM_IO and VM_PFNMAP frames

Handle VM_IO like VM_PFNMAP, as is common in the rest of Linux; extract
the formula to convert hva->pfn into a new function, which will soon
gain more capabilities.

Cc: Xiao Guangrong <guangrong.xiao@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: s390: inject PER i-fetch events on applicable icpts
David Hildenbrand [Tue, 24 May 2016 10:10:27 +0000 (12:10 +0200)]
KVM: s390: inject PER i-fetch events on applicable icpts

In case we have to emuluate an instruction or part of it (instruction,
partial instruction, operation exception), we have to inject a PER
instruction-fetching event for that instruction, if hardware told us to do
so.

In case we retry an instruction, we must not inject the PER event.

Please note that we don't filter the events properly yet, so guest
debugging will be visible for the guest.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: vmx: fix missed cancellation of TSC deadline timer
Wanpeng Li [Tue, 28 Jun 2016 06:54:19 +0000 (14:54 +0800)]
KVM: vmx: fix missed cancellation of TSC deadline timer

INFO: rcu_sched detected stalls on CPUs/tasks:
 1-...: (11800 GPs behind) idle=45d/140000000000000/0 softirq=0/0 fqs=21663
 (detected by 0, t=65016 jiffies, g=11500, c=11499, q=719)
Task dump for CPU 1:
qemu-system-x86 R  running task        0  3529   3525 0x00080808
 ffff8802021791a0 ffff880212895040 0000000000000001 00007f1c2c00db40
 ffff8801dd20fcd3 ffffc90002b98000 ffff8801dd20fc88 ffff8801dd20fcf8
 0000000000000286 ffff8801dd2ac538 ffff8801dd20fcc0 ffffffffc06949c9
Call Trace:
? kvm_write_guest_cached+0xb9/0x160 [kvm]
? __delay+0xf/0x20
? wait_lapic_expire+0x14a/0x200 [kvm]
? kvm_arch_vcpu_ioctl_run+0xcbe/0x1b00 [kvm]
? kvm_arch_vcpu_ioctl_run+0xe34/0x1b00 [kvm]
? kvm_vcpu_ioctl+0x2d3/0x7c0 [kvm]
? __fget+0x5/0x210
? do_vfs_ioctl+0x96/0x6a0
? __fget_light+0x2a/0x90
? SyS_ioctl+0x79/0x90
? do_syscall_64+0x7c/0x1e0
? entry_SYSCALL64_slow_path+0x25/0x25

This can be reproduced readily by running a full dynticks guest(since hrtimer
in guest is heavily used) w/ lapic_timer_advance disabled.

If fail to program hardware preemption timer, we will fallback to hrtimer based
method, however, a previous programmed preemption timer miss to cancel in this
scenario which results in one hardware preemption timer and one hrtimer emulated
tsc deadline timer run simultaneously. So sometimes the target guest deadline
tsc is earlier than guest tsc, which leads to the computation in vmx_set_hv_timer
can underflow and cause delta_tsc to be set a huge value, then host soft lockup
as above.

This patch fix it by cancelling the previous programmed preemption timer if there
is once we failed to program the new preemption timer and fallback to hrtimer
based method.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: x86: introduce cancel_hv_tscdeadline
Wanpeng Li [Thu, 30 Jun 2016 00:52:49 +0000 (08:52 +0800)]
KVM: x86: introduce cancel_hv_tscdeadline

Introduce cancel_hv_tscdeadline() to encapsulate preemption
timer cancel stuff.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: vmx: fix underflow in TSC deadline calculation
Paolo Bonzini [Mon, 27 Jun 2016 13:08:01 +0000 (15:08 +0200)]
KVM: vmx: fix underflow in TSC deadline calculation

If the TSC deadline timer is programmed really close to the deadline or
even in the past, the computation in vmx_set_hv_timer can underflow and
cause delta_tsc to be set to a huge value.  This generally results
in vmx_set_hv_timer returning -ERANGE, but we can fix it by limiting
delta_tsc to be positive or zero.

Reported-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: x86: use guest_exit_irqoff
Paolo Bonzini [Wed, 15 Jun 2016 13:23:11 +0000 (15:23 +0200)]
KVM: x86: use guest_exit_irqoff

This gains a few clock cycles per vmexit.  On Intel there is no need
anymore to enable the interrupts in vmx_handle_external_intr, since
we are using the "acknowledge interrupt on exit" feature.  AMD
needs to do that, and must be careful to avoid the interrupt shadow.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: x86: always use "acknowledge interrupt on exit"
Paolo Bonzini [Wed, 15 Jun 2016 18:55:08 +0000 (20:55 +0200)]
KVM: x86: always use "acknowledge interrupt on exit"

This is necessary to simplify handle_external_intr in the next patch.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: remove kvm_guest_enter/exit wrappers
Paolo Bonzini [Wed, 15 Jun 2016 13:18:26 +0000 (15:18 +0200)]
KVM: remove kvm_guest_enter/exit wrappers

Use the functions from context_tracking.h directly.

Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agocontext_tracking: move rcu_virt_note_context_switch out of kvm_host.h
Paolo Bonzini [Wed, 15 Jun 2016 13:09:28 +0000 (15:09 +0200)]
context_tracking: move rcu_virt_note_context_switch out of kvm_host.h

Make kvm_guest_{enter,exit} and __kvm_guest_{enter,exit} trivial wrappers
around the code in context_tracking.h.  Name the context_tracking.h functions
consistently with what those for kernel<->user switch.

Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMIPS: KVM: Combine entry trace events into class
James Hogan [Fri, 17 Jun 2016 17:19:40 +0000 (18:19 +0100)]
MIPS: KVM: Combine entry trace events into class

Combine the kvm_enter, kvm_reenter and kvm_out trace events into a
single kvm_transition event class to reduce duplication and bloat.

Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Fixes: 93258604ab6d ("MIPS: KVM: Add guest mode switch trace events")
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄ\8dmář <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agokvm: x86: use getboottime64
Arnd Bergmann [Fri, 17 Jun 2016 15:48:56 +0000 (17:48 +0200)]
kvm: x86: use getboottime64

KVM reads the current boottime value as a struct timespec in order to
calculate the guest wallclock time, resulting in an overflow in 2038
on 32-bit systems.

The data then gets passed as an unsigned 32-bit number to the guest,
and that in turn overflows in 2106.

We cannot do much about the second overflow, which affects both 32-bit
and 64-bit hosts, but we can ensure that they both behave the same
way and don't overflow until 2106, by using getboottime64() to read
a timespec64 value.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: VMX: enable guest access to LMCE related MSRs
Ashok Raj [Wed, 22 Jun 2016 06:59:56 +0000 (14:59 +0800)]
KVM: VMX: enable guest access to LMCE related MSRs

On Intel platforms, this patch adds LMCE to KVM MCE supported
capabilities and handles guest access to LMCE related MSRs.

Signed-off-by: Ashok Raj <ashok.raj@intel.com>
[Haozhong: macro KVM_MCE_CAP_SUPPORTED => variable kvm_mce_cap_supported
           Only enable LMCE on Intel platform
           Check MSR_IA32_FEATURE_CONTROL when handling guest
             access to MSR_IA32_MCG_EXT_CTL]
Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: VMX: validate individual bits of guest MSR_IA32_FEATURE_CONTROL
Haozhong Zhang [Wed, 22 Jun 2016 06:59:55 +0000 (14:59 +0800)]
KVM: VMX: validate individual bits of guest MSR_IA32_FEATURE_CONTROL

KVM currently does not check the value written to guest
MSR_IA32_FEATURE_CONTROL, though bits corresponding to disabled features
may be set. This patch makes KVM to validate individual bits written to
guest MSR_IA32_FEATURE_CONTROL according to enabled features.

Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoKVM: VMX: move msr_ia32_feature_control to vcpu_vmx
Haozhong Zhang [Wed, 22 Jun 2016 06:59:54 +0000 (14:59 +0800)]
KVM: VMX: move msr_ia32_feature_control to vcpu_vmx

msr_ia32_feature_control will be used for LMCE and not depend only on
nested anymore, so move it from struct nested_vmx to struct vcpu_vmx.

Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 years agoMerge tag 'kvm-s390-next-4.8-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 21 Jun 2016 13:21:51 +0000 (15:21 +0200)]
Merge tag 'kvm-s390-next-4.8-2' of git://git./linux/kernel/git/kvms390/linux into HEAD

KVM: s390: vSIE (nested virtualization) feature for 4.8 (kvm/next)

With an updated QEMU this allows to create nested KVM guests
(KVM under KVM) on s390.

s390 memory management changes from Martin Schwidefsky or
acked by Martin. One common code memory management change (pageref)
acked by Andrew Morton.

The feature has to be enabled with the nested medule parameter.

8 years agoKVM: s390: vsie: add module parameter "nested"
David Hildenbrand [Tue, 2 Feb 2016 14:41:22 +0000 (15:41 +0100)]
KVM: s390: vsie: add module parameter "nested"

Let's be careful first and allow nested virtualization only if enabled
by the system administrator. In addition, user space still has to
explicitly enable it via SCLP features for it to work.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: add indication for future features
David Hildenbrand [Wed, 13 Apr 2016 15:06:50 +0000 (17:06 +0200)]
KVM: s390: vsie: add indication for future features

We have certain SIE features that we cannot support for now.
Let's add these features, so user space can directly prepare to enable
them, so we don't have to update yet another component.

In addition, add a comment block, telling why it is for now not possible to
forward/enable these features.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: correctly set and handle guest TOD
David Hildenbrand [Thu, 29 Oct 2015 09:30:36 +0000 (10:30 +0100)]
KVM: s390: vsie: correctly set and handle guest TOD

Guest 2 sets up the epoch of guest 3 from his point of view. Therefore,
we have to add the guest 2 epoch to the guest 3 epoch. We also have to take
care of guest 2 epoch changes on STP syncs. This will work just fine by
also updating the guest 3 epoch when a vsie_block has been set for a VCPU.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: speed up VCPU external calls
David Hildenbrand [Tue, 7 Jul 2015 18:39:35 +0000 (20:39 +0200)]
KVM: s390: vsie: speed up VCPU external calls

Whenever a SIGP external call is injected via the SIGP external call
interpretation facility, the VCPU is not kicked. When a VCPU is currently
in the VSIE, the external call might not be processed immediately.

Therefore we have to provoke partial execution exceptions, which leads to a
kick of the VCPU and therefore also kick out of VSIE. This is done by
simulating the WAIT state. This bit has no other side effects.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: don't use CPUSTAT_WAIT to detect if a VCPU is idle
David Hildenbrand [Thu, 18 Feb 2016 09:15:43 +0000 (10:15 +0100)]
KVM: s390: don't use CPUSTAT_WAIT to detect if a VCPU is idle

As we want to make use of CPUSTAT_WAIT also when a VCPU is not idle but
to force interception of external calls, let's check in the bitmap instead.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: speed up VCPU irq delivery when handling vsie
David Hildenbrand [Fri, 27 May 2016 20:03:52 +0000 (22:03 +0200)]
KVM: s390: vsie: speed up VCPU irq delivery when handling vsie

Whenever we want to wake up a VCPU (e.g. when injecting an IRQ), we
have to kick it out of vsie, so the request will be handled faster.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: try to refault after a reported fault to g2
David Hildenbrand [Wed, 8 Jul 2015 11:25:31 +0000 (13:25 +0200)]
KVM: s390: vsie: try to refault after a reported fault to g2

We can avoid one unneeded SIE entry after we reported a fault to g2.
Theoretically, g2 resolves the fault and we can create the shadow mapping
directly, instead of failing again when entering the SIE.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support IBS interpretation
David Hildenbrand [Tue, 24 Nov 2015 15:56:23 +0000 (16:56 +0100)]
KVM: s390: vsie: support IBS interpretation

We can easily enable ibs for guest 2, so he can use it for guest 3.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support conditional-external-interception
David Hildenbrand [Tue, 24 Nov 2015 15:54:37 +0000 (16:54 +0100)]
KVM: s390: vsie: support conditional-external-interception

We can easily enable cei for guest 2, so he can use it for guest 3.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support intervention-bypass
David Hildenbrand [Tue, 24 Nov 2015 15:53:51 +0000 (16:53 +0100)]
KVM: s390: vsie: support intervention-bypass

We can easily enable intervention bypass for guest 2, so it can use it
for guest 3.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support guest-storage-limit-suppression
David Hildenbrand [Tue, 24 Nov 2015 15:41:33 +0000 (16:41 +0100)]
KVM: s390: vsie: support guest-storage-limit-suppression

We can easily forward guest-storage-limit-suppression if available.

One thing to care about is keeping the prefix properly mapped when
gsls in toggled on/off or the mso changes in between. Therefore we better
remap the prefix on any mso changes just like we already do with the
prefix.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support guest-PER-enhancement
David Hildenbrand [Tue, 24 Nov 2015 15:32:35 +0000 (16:32 +0100)]
KVM: s390: vsie: support guest-PER-enhancement

We can easily forward the guest-PER-enhancement facility to guest 2 if
available.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support shared IPTE-interlock facility
David Hildenbrand [Wed, 25 Nov 2015 08:59:49 +0000 (09:59 +0100)]
KVM: s390: vsie: support shared IPTE-interlock facility

As we forward the whole SCA provided by guest 2, we can directly forward
SIIF if available.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support 64-bit-SCAO
David Hildenbrand [Wed, 25 Nov 2015 10:02:26 +0000 (11:02 +0100)]
KVM: s390: vsie: support 64-bit-SCAO

Let's provide the 64-bit-SCAO facility to guest 2, so he can set up a SCA
for guest 3 that has a 64 bit address. Please note that we already require
the 64 bit SCAO for our vsie implementation, in order to forward the SCA
directly (by pinning the page).

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support run-time-instrumentation
David Hildenbrand [Tue, 26 Jan 2016 11:51:06 +0000 (12:51 +0100)]
KVM: s390: vsie: support run-time-instrumentation

As soon as guest 2 is allowed to use run-time-instrumentation (indicated
via via STFLE), it can also enable it for guest 3.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support vectory facility (SIMD)
David Hildenbrand [Wed, 25 Nov 2015 10:08:32 +0000 (11:08 +0100)]
KVM: s390: vsie: support vectory facility (SIMD)

As soon as guest 2 is allowed to use the vector facility (indicated via
STFLE), it can also enable it for guest 3. We have to take care of the
sattellite block that might be used when not relying on lazy vector
copying (not the case for KVM).

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support transactional execution
David Hildenbrand [Wed, 25 Nov 2015 10:13:32 +0000 (11:13 +0100)]
KVM: s390: vsie: support transactional execution

As soon as guest 2 is allowed to use transactional execution (indicated via
STFLE), he can also enable it for guest 3.

Active transactional execution requires also the second prefix page to be
mapped. If that page cannot be mapped, a validity icpt has to be presented
to the guest.

We have to take care of tx being toggled on/off, otherwise we might get
wrong prefix validity icpt.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support aes dea wrapping keys
David Hildenbrand [Thu, 26 Nov 2015 12:11:42 +0000 (13:11 +0100)]
KVM: s390: vsie: support aes dea wrapping keys

As soon as message-security-assist extension 3 is enabled for guest 2,
we have to allow key wrapping for guest 3.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support STFLE interpretation
David Hildenbrand [Thu, 26 Nov 2015 13:11:19 +0000 (14:11 +0100)]
KVM: s390: vsie: support STFLE interpretation

Issuing STFLE is extremely rare. Instead of copying 2k on every
VSIE call, let's do this lazily, when a guest 3 tries to execute
STFLE. We can setup the block and retry.

Unfortunately, we can't directly forward that facility list, as
we only have a 31 bit address for the facility list designation.
So let's use a DMA allocation for our vsie_page instead for now.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support host-protection-interruption
David Hildenbrand [Fri, 27 Nov 2015 11:34:28 +0000 (12:34 +0100)]
KVM: s390: vsie: support host-protection-interruption

Introduced with ESOP, therefore available for the guest if it
is allowed to use ESOP.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support edat1 / edat2
David Hildenbrand [Fri, 12 Feb 2016 11:24:20 +0000 (12:24 +0100)]
KVM: s390: vsie: support edat1 / edat2

If guest 2 is allowed to use edat 1 / edat 2, it can also set it up for
guest 3, so let's properly check and forward the edat cpuflags.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: support setting the ibc
David Hildenbrand [Fri, 19 Feb 2016 09:11:24 +0000 (10:11 +0100)]
KVM: s390: vsie: support setting the ibc

As soon as we forward an ibc to guest 2 (indicated via
kvm->arch.model.ibc), he can also use it for guest 3. Let's properly round
the ibc up/down, so we avoid any potential validity icpts from the
underlying SIE, if it doesn't simply round the values.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: optimize gmap prefix mapping
David Hildenbrand [Fri, 22 Apr 2016 11:50:09 +0000 (13:50 +0200)]
KVM: s390: vsie: optimize gmap prefix mapping

In order to not always map the prefix, we have to take care of certain
aspects that implicitly unmap the prefix:
- Changes to the prefix address
- Changes to MSO, because the HVA of the prefix is changed
- Changes of the gmap shadow (e.g. unshadowed, asce or edat changes)

By properly handling these cases, we can stop remapping the prefix when
there is no reason to do so.

This also allows us now to not acquire any gmap shadow locks when
rerunning the vsie and still having a valid gmap shadow.

Please note, to detect changing gmap shadows, we have to keep the reference
of the gmap shadow. The address of a gmap shadow does otherwise not
reliably indicate if the gmap shadow has changed (the memory chunk
could get reused).

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: vsie: initial support for nested virtualization
David Hildenbrand [Wed, 8 Jul 2015 11:19:48 +0000 (13:19 +0200)]
KVM: s390: vsie: initial support for nested virtualization

This patch adds basic support for nested virtualization on s390x, called
VSIE (virtual SIE) and allows it to be used by the guest if the necessary
facilities are supported by the hardware and enabled for the guest.

In order to make this work, we have to shadow the sie control block
provided by guest 2. In order to gain some performance, we have to
reuse the same shadow blocks as good as possible. For now, we allow
as many shadow blocks as we have VCPUs (that way, every VCPU can run the
VSIE concurrently).

We have to watch out for the prefix getting unmapped out of our shadow
gmap and properly get the VCPU out of VSIE in that case, to fault the
prefix pages back in. We use the PROG_REQUEST bit for that purpose.

This patch is based on an initial prototype by Tobias Elpelt.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agomm/page_ref: introduce page_ref_inc_return
David Hildenbrand [Mon, 20 Jun 2016 10:09:41 +0000 (12:09 +0200)]
mm/page_ref: introduce page_ref_inc_return

Let's introduce that helper.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390: introduce page_to_virt() and pfn_to_virt()
David Hildenbrand [Mon, 16 Nov 2015 09:45:03 +0000 (10:45 +0100)]
s390: introduce page_to_virt() and pfn_to_virt()

Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: backup the currently enabled gmap when scheduled out
David Hildenbrand [Wed, 11 Mar 2015 15:47:33 +0000 (16:47 +0100)]
KVM: s390: backup the currently enabled gmap when scheduled out

Nested virtualization will have to enable own gmaps. Current code
would enable the wrong gmap whenever scheduled out and back in,
therefore resulting in the wrong gmap being enabled.

This patch reenables the last enabled gmap, therefore avoiding having to
touch vcpu->arch.gmap when enabling a different gmap.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: s390: fast path for shadow gmaps in gmap notifier
David Hildenbrand [Mon, 27 Apr 2015 14:29:34 +0000 (16:29 +0200)]
KVM: s390: fast path for shadow gmaps in gmap notifier

The default kvm gmap notifier doesn't have to handle shadow gmaps.
So let's just directly exit in case we get notified about one.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: don't fault everything in read-write in gmap_pte_op_fixup()
David Hildenbrand [Mon, 13 Jun 2016 08:49:04 +0000 (10:49 +0200)]
s390/mm: don't fault everything in read-write in gmap_pte_op_fixup()

Let's not fault in everything in read-write but limit it to read-only
where possible.

When restricting access rights, we already have the required protection
level in our hands. When reading from guest 2 storage (gmap_read_table),
it is obviously PROT_READ. When shadowing a pte, the required protection
level is given via the guest 2 provided pte.

Based on an initial patch by Martin Schwidefsky.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: allow to check if a gmap shadow is valid
David Hildenbrand [Fri, 27 May 2016 16:57:33 +0000 (18:57 +0200)]
s390/mm: allow to check if a gmap shadow is valid

It will be very helpful to have a mechanism to check without any locks
if a given gmap shadow is still valid and matches the given properties.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: remember the int code for the last gmap fault
David Hildenbrand [Tue, 8 Mar 2016 11:31:52 +0000 (12:31 +0100)]
s390/mm: remember the int code for the last gmap fault

For nested virtualization, we want to know if we are handling a protection
exception, because these can directly be forwarded to the guest without
additional checks.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: limit number of real-space gmap shadows
David Hildenbrand [Mon, 2 May 2016 10:10:17 +0000 (12:10 +0200)]
s390/mm: limit number of real-space gmap shadows

We have no known user of real-space designation and only support it to
be architecture compliant.

Gmap shadows with real-space designation are never unshadowed
automatically, as there is nothing to protect for the top level table.

So let's simply limit the number of such shadows to one by removing
existing ones on creation of another one.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: support real-space for gmap shadows
David Hildenbrand [Mon, 18 Apr 2016 14:22:24 +0000 (16:22 +0200)]
s390/mm: support real-space for gmap shadows

We can easily support real-space designation just like EDAT1 and EDAT2.
So guest2 can provide for guest3 an asce with the real-space control being
set.

We simply have to allocate the biggest page table possible and fake all
levels.

There is no protection to consider. If we exceed guest memory, vsie code
will inject an addressing exception (via program intercept). In the future,
we could limit the fake table level to the gmap page table.

As the top level page table can never go away, such gmap shadows will never
get unshadowed, we'll have to come up with another way to limit the number
of kept gmap shadows.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: push rte protection down to shadow pte
David Hildenbrand [Mon, 18 Apr 2016 15:46:21 +0000 (17:46 +0200)]
s390/mm: push rte protection down to shadow pte

Just like we already do with ste protection, let's take rte protection
into account. This way, the host pte doesn't have to be mapped writable.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: support EDAT2 for gmap shadows
David Hildenbrand [Mon, 18 Apr 2016 11:42:05 +0000 (13:42 +0200)]
s390/mm: support EDAT2 for gmap shadows

If the guest is enabled for EDAT2, we can easily create shadows for
guest2 -> guest3 provided tables that make use of EDAT2.

If guest2 references a 2GB page, this memory looks consecutive for guest2,
but it does not have to be so for us. Therefore we have to create fake
segment and page tables.

This works just like EDAT1 support, so page tables are removed when the
parent table (r3t table entry) is changed.

We don't hve to care about:
- ACCF-Validity Control in RTTE
- Access-Control Bits in RTTE
- Fetch-Protection Bit in RTTE
- Common-Region Bit in RTTE

Just like for EDAT1, all bits might be dropped and there is no guaranteed
that they are active.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: support EDAT1 for gmap shadows
David Hildenbrand [Mon, 18 Apr 2016 11:24:52 +0000 (13:24 +0200)]
s390/mm: support EDAT1 for gmap shadows

If the guest is enabled for EDAT1, we can easily create shadows for
guest2 -> guest3 provided tables that make use of EDAT1.

If guest2 references a 1MB page, this memory looks consecutive for guest2,
but it might not be so for us. Therefore we have to create fake page tables.

We can easily add that to our existing infrastructure. The invalidation
mechanism will make sure that fake page tables are removed when the parent
table (sgt table entry) is changed.

As EDAT1 also introduced protection on all page table levels, we have to
also shadow these correctly.

We don't have to care about:
- ACCF-Validity Control in STE
- Access-Control Bits in STE
- Fetch-Protection Bit in STE
- Common-Segment Bit in STE

As all bits might be dropped and there is no guaranteed that they are
active ("unpredictable whether the CPU uses these bits", "may be used").
Without using EDAT1 in the shadow ourselfes (STE-format control == 0),
simply shadowing these bits would not be enough. They would be ignored.

Please note that we are using the "fake" flag to make this look consistent
with further changes (EDAT2, real-space designation support) and don't let
the shadow functions handle fc=1 stes.

In the future, with huge pages in the host, gmap_shadow_pgt() could simply
try to map a huge host page if "fake" is set to one and indicate via return
value that no lower fake tables / shadow ptes are required.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: prepare for EDAT1/EDAT2 support in gmap shadow
David Hildenbrand [Tue, 8 Mar 2016 11:17:40 +0000 (12:17 +0100)]
s390/mm: prepare for EDAT1/EDAT2 support in gmap shadow

In preparation for EDAT1/EDAT2 support for gmap shadows, we have to store
the requested edat level in the gmap shadow.

The edat level used during shadow translation is a property of the gmap
shadow. Depending on that level, the gmap shadow will look differently for
the same guest tables. We have to store it internally in order to support
it later.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: push ste protection down to shadow pte
David Hildenbrand [Mon, 18 Apr 2016 15:19:59 +0000 (17:19 +0200)]
s390/mm: push ste protection down to shadow pte

If a guest ste is read-only, it doesn't make sense to force the ptes in as
writable in the host. If the source page is read-only in the host, it won't
have to be made writable. Please note that if the source page is not
available, it will still be faulted in writable. This can be changed
internally later on.

If ste protection is removed, underlying shadow tables are also removed,
therefore this change does not affect the guest.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: take ipte_lock during shadow faults
David Hildenbrand [Wed, 27 Jan 2016 16:24:03 +0000 (17:24 +0100)]
s390/mm: take ipte_lock during shadow faults

Let's take the ipte_lock while working on guest 2 provided page table, just
like the other gaccess functions.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: protection exceptions are corrrectly shadowed
David Hildenbrand [Wed, 27 Jan 2016 16:18:41 +0000 (17:18 +0100)]
s390/mm: protection exceptions are corrrectly shadowed

As gmap shadows contains correct protection permissions, protection
exceptons can directly be forwarded to guest 3. If we would encounter
a protection exception while faulting, the next guest 3 run will
automatically handle that for us.

Keep the dat_protection logic in place, as it will be helpful later.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: take the mmap_sem in kvm_s390_shadow_fault()
David Hildenbrand [Tue, 2 Feb 2016 11:26:00 +0000 (12:26 +0100)]
s390/mm: take the mmap_sem in kvm_s390_shadow_fault()

Instead of doing it in the caller, let's just take the mmap_sem
in kvm_s390_shadow_fault(). By taking it as read, we allow parallel
faulting on shadow page tables, gmap shadow code is prepared for that.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: fix races on gmap_shadow creation
David Hildenbrand [Tue, 8 Mar 2016 11:30:46 +0000 (12:30 +0100)]
s390/mm: fix races on gmap_shadow creation

Before any thread is allowed to use a gmap_shadow, it has to be fully
initialized. However, for invalidation to work properly, we have to
register the new gmap_shadow before we protect the parent gmap table.

Because locking is tricky, and we have to avoid duplicate gmaps, let's
introduce an initialized field, that signalizes other threads if that
gmap_shadow can already be used or if they have to retry.

Let's properly return errors using ERR_PTR() instead of simply returning
NULL, so a caller can properly react on the error.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: avoid races on region/segment/page table shadowing
David Hildenbrand [Tue, 8 Mar 2016 11:23:38 +0000 (12:23 +0100)]
s390/mm: avoid races on region/segment/page table shadowing

We have to unlock sg->guest_table_lock in order to call
gmap_protect_rmap(). If we sleep just before that call, another VCPU
might pick up that shadowed page table (while it is not protected yet)
and use it.

In order to avoid these races, we have to introduce a third state -
"origin set but still invalid" for an entry. This way, we can avoid
another thread already using the entry before the table is fully protected.
As soon as everything is set up, we can clear the invalid bit - if we
had no race with the unshadowing code.

Suggested-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: shadow pages with real guest requested protection
David Hildenbrand [Tue, 8 Mar 2016 11:21:41 +0000 (12:21 +0100)]
s390/mm: shadow pages with real guest requested protection

We really want to avoid manually handling protection for nested
virtualization. By shadowing pages with the protection the guest asked us
for, the SIE can handle most protection-related actions for us (e.g.
special handling for MVPG) and we can directly forward protection
exceptions to the guest.

PTEs will now always be shadowed with the correct _PAGE_PROTECT flag.
Unshadowing will take care of any guest changes to the parent PTE and
any host changes to the host PTE. If the host PTE doesn't have the
fitting access rights or is not available, we have to fix it up.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: flush tlb of shadows in all situations
David Hildenbrand [Fri, 15 Apr 2016 10:45:45 +0000 (12:45 +0200)]
s390/mm: flush tlb of shadows in all situations

For now, the tlb of shadow gmap is only flushed when the parent is removed,
not when it is removed upfront. Therefore other shadow gmaps can reuse the
tables without the tlb getting flushed.

Fix this by simply flushing the tlb
1. Before the shadow tables are removed (analogouos to other unshadow functions)
2. When the gmap is freed and therefore the top level pages are freed.

Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: add kvm shadow fault function
Martin Schwidefsky [Tue, 8 Mar 2016 11:16:35 +0000 (12:16 +0100)]
s390/mm: add kvm shadow fault function

This patch introduces function kvm_s390_shadow_fault() used to resolve a
fault on a shadow gmap. This function will do validity checking and
build up the shadow page table hierarchy in order to fault in the
requested page into the shadow page table structure.

If an exception occurs while shadowing, guest 2 has to be notified about
it using either an exception or a program interrupt intercept. If
concurrent unshadowing occurres, this function will simply return with
-EAGAIN and the caller has to retry.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: add shadow gmap support
Martin Schwidefsky [Tue, 8 Mar 2016 11:12:18 +0000 (12:12 +0100)]
s390/mm: add shadow gmap support

For a nested KVM guest the outer KVM host needs to create shadow
page tables for the nested guest. This patch adds the basic support
to the guest address space (gmap) code.

For each guest address space the inner KVM host creates, the first
outer KVM host needs to create shadow page tables. The address space
is identified by the ASCE loaded into the control register 1 at the
time the inner SIE instruction for the second nested KVM guest is
executed. The outer KVM host creates the shadow tables starting with
the table identified by the ASCE on a on-demand basis. The outer KVM
host will get repeated faults for all the shadow tables needed to
run the second KVM guest.

While a shadow page table for the second KVM guest is active the access
to the origin region, segment and page tables needs to be restricted
for the first KVM guest. For region and segment and page tables the first
KVM guest may read the memory, but write attempt has to lead to an
unshadow.  This is done using the page invalid and read-only bits in the
page table of the first KVM guest. If the first guest re-accesses one of
the origin pages of a shadow, it gets a fault and the affected parts of
the shadow page table hierarchy needs to be removed again.

PGSTE tables don't have to be shadowed, as all interpretation assist can't
deal with the invalid bits in the shadow pte being set differently than
the original ones provided by the first KVM guest.

Many bug fixes and improvements by David Hildenbrand.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: add reference counter to gmap structure
Martin Schwidefsky [Tue, 8 Mar 2016 10:55:04 +0000 (11:55 +0100)]
s390/mm: add reference counter to gmap structure

Let's use a reference counter mechanism to control the lifetime of
gmap structures. This will be needed for further changes related to
gmap shadows.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: extended gmap pte notifier
Martin Schwidefsky [Tue, 8 Mar 2016 10:54:42 +0000 (11:54 +0100)]
s390/mm: extended gmap pte notifier

The current gmap pte notifier forces a pte into to a read-write state.
If the pte is invalidated the gmap notifier is called to inform KVM
that the mapping will go away.

Extend this approach to allow read-write, read-only and no-access
as possible target states and call the pte notifier for any change
to the pte.

This mechanism is used to temporarily set specific access rights for
a pte without doing the heavy work of a true mprotect call.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/mm: use RCU for gmap notifier list and the per-mm gmap list
Martin Schwidefsky [Tue, 8 Mar 2016 10:54:14 +0000 (11:54 +0100)]
s390/mm: use RCU for gmap notifier list and the per-mm gmap list

The gmap notifier list and the gmap list in the mm_struct change rarely.
Use RCU to optimize the reader of these lists.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agos390/kvm: page table invalidation notifier
Martin Schwidefsky [Tue, 8 Mar 2016 10:52:54 +0000 (11:52 +0100)]
s390/kvm: page table invalidation notifier

Pass an address range to the page table invalidation notifier
for KVM. This allows to notify changes that affect a larger
virtual memory area, e.g. for 1MB pages.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
8 years agoKVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
Mahesh Salgaonkar [Sun, 15 May 2016 04:14:26 +0000 (09:44 +0530)]
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt

When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.

When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.

If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.

With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.

This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.

On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.

All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.

It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.

One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete.  Once TB resync is complete all threads will then
proceed.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
8 years agopowerpc/powernv: Remove the usage of PACAR1 from opal wrappers
Mahesh Salgaonkar [Sun, 15 May 2016 04:14:13 +0000 (09:44 +0530)]
powerpc/powernv: Remove the usage of PACAR1 from opal wrappers

OPAL_CALL wrapper code sticks the r1 (stack pointer) into PACAR1 purely
for debugging purpose only. The power7_wakeup* functions relies on stack
pointer saved in PACAR1. Any opal call made using opal wrapper (directly
or in-directly) before we fall through power7_wakeup*, then it ends up
replacing r1 in PACAR1(r13) leading to kernel panic. So far we don't see
any issues because we have never made any opal calls using OPAL wrapper
before power7_wakeup*. But the subsequent HMI patch would need to invoke
C calls during cpu wakeup/idle path that in-directly makes opal call using
opal wrapper. This patch facilitates the subsequent HMI patch by removing
usage of PACAR1 from opal call wrapper.

Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
8 years agoKVM: PPC: Book3S PR: Fix contents of SRR1 when injecting a program exception
Thomas Huth [Thu, 19 May 2016 09:33:31 +0000 (11:33 +0200)]
KVM: PPC: Book3S PR: Fix contents of SRR1 when injecting a program exception

vcpu->arch.shadow_srr1 only contains usable values for injecting
a program exception into the guest if we entered the function
kvmppc_handle_exit_pr() with exit_nr == BOOK3S_INTERRUPT_PROGRAM.
In other cases, the shadow_srr1 bits are zero. Since we want to
pass an illegal-instruction program check to the guest, set
"flags" to SRR1_PROGILL for these other cases.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>