GitHub/moto-9609/android_kernel_motorola_exynos9610.git
14 years agoKVM: MMU: fix page dirty tracking lost while sync page
Xiao Guangrong [Fri, 16 Jul 2010 03:25:17 +0000 (11:25 +0800)]
KVM: MMU: fix page dirty tracking lost while sync page

In sync-page path, if spte.writable is changed, it will lose page dirty
tracking, for example:

assume spte.writable = 0 in a unsync-page, when it's synced, it map spte
to writable(that is spte.writable = 1), later guest write spte.gfn, it means
spte.gfn is dirty, then guest changed this mapping to read-only, after it's
synced,  spte.writable = 0

So, when host release the spte, it detect spte.writable = 0 and not mark page
dirty

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: fix broken page accessed tracking with ept enabled
Xiao Guangrong [Fri, 16 Jul 2010 03:23:04 +0000 (11:23 +0800)]
KVM: MMU: fix broken page accessed tracking with ept enabled

In current code, if ept is enabled(shadow_accessed_mask = 0), the page
accessed tracking is lost.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: add missing reserved bits check in speculative path
Xiao Guangrong [Fri, 16 Jul 2010 03:19:51 +0000 (11:19 +0800)]
KVM: MMU: add missing reserved bits check in speculative path

In the speculative path, we should check guest pte's reserved bits just as
the real processor does

Reported-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: fix mmu notifier invalidate handler for huge spte
Andrea Arcangeli [Fri, 16 Jul 2010 09:52:55 +0000 (11:52 +0200)]
KVM: MMU: fix mmu notifier invalidate handler for huge spte

The index wasn't calculated correctly (off by one) for huge spte so KVM guest
was unstable with transparent hugepages.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix xchg instruction emulation
Wei Yongjun [Thu, 15 Jul 2010 00:51:58 +0000 (08:51 +0800)]
KVM: x86 emulator: fix xchg instruction emulation

If the destination is a memory operand and the memory cannot
map to a valid page, the xchg instruction emulation and locked
instruction will not work on io regions and stuck in endless
loop. We should emulate exchange as write to fix it.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Acked-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: Call mask notifiers from pic
Gleb Natapov [Thu, 15 Jul 2010 09:24:37 +0000 (12:24 +0300)]
KVM: x86: Call mask notifiers from pic

If pit delivers interrupt while pic is masking it OS will never do EOI
and ack notifier will not be called so when pit will be unmasked no pit
interrupts will be delivered any more. Calling mask notifiers solves this
issue.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: never re-execute instruction with enabled tdp
Gleb Natapov [Wed, 14 Jul 2010 16:05:45 +0000 (19:05 +0300)]
KVM: x86: never re-execute instruction with enabled tdp

With tdp enabled we should get into emulator only when emulating io, so
reexecution will always bring us back into emulator.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Document KVM_GET_SUPPORTED_CPUID2 ioctl
Avi Kivity [Wed, 14 Jul 2010 06:45:21 +0000 (09:45 +0300)]
KVM: Document KVM_GET_SUPPORTED_CPUID2 ioctl

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: emulator: inc/dec can have lock prefix
Gleb Natapov [Tue, 13 Jul 2010 13:40:23 +0000 (16:40 +0300)]
KVM: x86: emulator: inc/dec can have lock prefix

Mark inc (0xfe/0 0xff/0) and dec (0xfe/1 0xff/1) as lock prefix capable.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Eliminate redundant temporaries in FNAME(fetch)
Avi Kivity [Tue, 13 Jul 2010 11:27:11 +0000 (14:27 +0300)]
KVM: MMU: Eliminate redundant temporaries in FNAME(fetch)

'level' and 'sptep' are aliases for 'interator.level' and 'iterator.sptep', no
need for them.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Validate all gptes during fetch, not just those used for new pages
Avi Kivity [Tue, 13 Jul 2010 11:27:10 +0000 (14:27 +0300)]
KVM: MMU: Validate all gptes during fetch, not just those used for new pages

Currently, when we fetch an spte, we only verify that gptes match those that
the walker saw if we build new shadow pages for them.

However, this misses the following race:

  vcpu1            vcpu2

  walk
                  change gpte
                  walk
                  instantiate sp

  fetch existing sp

Fix by validating every gpte, regardless of whether it is used for building
a new sp or not.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Simplify spte fetch() function
Avi Kivity [Tue, 13 Jul 2010 11:27:09 +0000 (14:27 +0300)]
KVM: MMU: Simplify spte fetch() function

Partition the function into three sections:

- fetching indirect shadow pages (host_level > guest_level)
- fetching direct shadow pages (page_level < host_level <= guest_level)
- the final spte (page_level == host_level)

Instead of the current spaghetti.

A slight change from the original code is that we call validate_direct_spte()
more often: previously we called it only for gw->level, now we also call it for
lower levels.  The change should have no effect.

[xiao: fix regression caused by validate_direct_spte() called too late]

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Add gpte_valid() helper
Avi Kivity [Tue, 13 Jul 2010 11:27:08 +0000 (14:27 +0300)]
KVM: MMU: Add gpte_valid() helper

Move the code to check whether a gpte has changed since we fetched it into
a helper.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Add validate_direct_spte() helper
Avi Kivity [Tue, 13 Jul 2010 11:27:07 +0000 (14:27 +0300)]
KVM: MMU: Add validate_direct_spte() helper

Add a helper to verify that a direct shadow page is valid wrt the required
access permissions; drop the page if it is not valid.

Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Add drop_large_spte() helper
Avi Kivity [Tue, 13 Jul 2010 11:27:06 +0000 (14:27 +0300)]
KVM: MMU: Add drop_large_spte() helper

To clarify spte fetching code, move large spte handling into a helper.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Use __set_spte to link shadow pages
Avi Kivity [Tue, 13 Jul 2010 11:27:05 +0000 (14:27 +0300)]
KVM: MMU: Use __set_spte to link shadow pages

To avoid split accesses to 64 bit sptes on i386, use __set_spte() to link
shadow pages together.

(not technically required since shadow pages are __GFP_KERNEL, so upper 32
bits are always clear)

Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: Add link_shadow_page() helper
Avi Kivity [Tue, 13 Jul 2010 11:27:04 +0000 (14:27 +0300)]
KVM: MMU: Add link_shadow_page() helper

To simplify the process of fetching an spte, add a helper that links
a shadow page to an spte.

Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Convert mask notifiers to use irqchip/pin instead of gsi
Gleb Natapov [Sun, 11 Jul 2010 12:32:23 +0000 (15:32 +0300)]
KVM: Convert mask notifiers to use irqchip/pin instead of gsi

Devices register mask notifier using gsi, but irqchip knows about
irqchip/pin, so conversion from irqchip/pin to gsi should be done before
looking for mask notifier to call.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Document MCE banks non-exposure via KVM_GET_MSR_INDEX_LIST
Avi Kivity [Wed, 7 Jul 2010 11:09:39 +0000 (14:09 +0300)]
KVM: Document MCE banks non-exposure via KVM_GET_MSR_INDEX_LIST

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Expose MCE control MSRs to userspace
Avi Kivity [Wed, 7 Jul 2010 11:09:38 +0000 (14:09 +0300)]
KVM: Expose MCE control MSRs to userspace

Userspace needs to reset and save/restore these MSRs.

The MCE banks are not exposed since their number varies from vcpu to vcpu.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: PIT: stop vpit before freeing irq_routing
Xiao Guangrong [Sat, 10 Jul 2010 09:37:56 +0000 (17:37 +0800)]
KVM: PIT: stop vpit before freeing irq_routing

Fix:
general protection fault: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
......
Call Trace:
 [<ffffffffa0159bd1>] ? kvm_set_irq+0xdd/0x24b [kvm]
 [<ffffffff8106ea8b>] ? trace_hardirqs_off_caller+0x1f/0x10e
 [<ffffffff813ad17f>] ? sub_preempt_count+0xe/0xb6
 [<ffffffff8106d273>] ? put_lock_stats+0xe/0x27
...
RIP  [<ffffffffa0159c72>] kvm_set_irq+0x17e/0x24b [kvm]

This bug is triggered when guest is shutdown, is because we freed
irq_routing before pit thread stopped

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Reenter guest after emulation failure if due to access to non-mmio address
Gleb Natapov [Thu, 8 Jul 2010 09:41:12 +0000 (12:41 +0300)]
KVM: Reenter guest after emulation failure if due to access to non-mmio address

When shadow pages are in use sometimes KVM try to emulate an instruction
when it accesses a shadowed page. If emulation fails KVM un-shadows the
page and reenter guest to allow vcpu to execute the instruction. If page
is not in shadow page hash KVM assumes that this was attempt to do MMIO
and reports emulation failure to userspace since there is no way to fix
the situation. This logic has a race though. If two vcpus tries to write
to the same shadowed page simultaneously both will enter emulator, but
only one of them will find the page in shadow page hash since the one who
founds it also removes it from there, so another cpu will report failure
to userspace and will abort the guest.

Fix this by checking (in addition to checking shadowed page hash) that
page that caused the emulation belongs to valid memory slot. If it is
then reenter the guest to allow vcpu to reexecute the instruction.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Return EFAULT from kvm ioctl when guest accesses bad area
Gleb Natapov [Wed, 7 Jul 2010 17:16:45 +0000 (20:16 +0300)]
KVM: Return EFAULT from kvm ioctl when guest accesses bad area

Currently if guest access address that belongs to memory slot but is not
backed up by page or page is read only KVM treats it like MMIO access.
Remove that capability. It was never part of the interface and should
not be relied upon.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: define hwpoison variables static
Gleb Natapov [Wed, 7 Jul 2010 17:16:44 +0000 (20:16 +0300)]
KVM: define hwpoison variables static

They are not used outside of the file.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: fix lock imbalance in kvm_create_pit()
Jiri Slaby [Wed, 7 Jul 2010 13:02:25 +0000 (15:02 +0200)]
KVM: fix lock imbalance in kvm_create_pit()

Stanse found that there is an omitted unlock in kvm_create_pit in one fail
path. Add proper unlock there.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: x86@kernel.org
Cc: Gleb Natapov <gleb@redhat.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Gregory Haskins <ghaskins@novell.com>
Cc: kvm@vger.kernel.org
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: Keep going on permission error
Avi Kivity [Tue, 6 Jul 2010 13:20:43 +0000 (16:20 +0300)]
KVM: MMU: Keep going on permission error

Real hardware disregards permission errors when computing page fault error
code bit 0 (page present).  Do the same.

Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: Only indicate a fetch fault in page fault error code if nx is enabled
Avi Kivity [Tue, 6 Jul 2010 12:40:18 +0000 (15:40 +0300)]
KVM: MMU: Only indicate a fetch fault in page fault error code if nx is enabled

Bit 4 of the page fault error code is set only if EFER.NX is set.

Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: re-implementing 'mov AL,moffs' instruction decoding
Wei Yongjun [Wed, 7 Jul 2010 09:43:35 +0000 (17:43 +0800)]
KVM: x86 emulator: re-implementing 'mov AL,moffs' instruction decoding

This patch change to use DstAcc for decoding 'mov AL, moffs'
and introduced SrcAcc for decoding 'mov moffs, AL'.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix cli/sti instruction emulation
Wei Yongjun [Tue, 6 Jul 2010 08:54:19 +0000 (16:54 +0800)]
KVM: x86 emulator: fix cli/sti instruction emulation

If IOPL check fail, the cli/sti emulate GP and then we should
skip writeback since the default write OP is OP_REG.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix 'mov rm,sreg' instruction decoding
Wei Yongjun [Tue, 6 Jul 2010 08:52:53 +0000 (16:52 +0800)]
KVM: x86 emulator: fix 'mov rm,sreg' instruction decoding

The source operand of 'mov rm,sreg' is segment register, not
general-purpose register, so remove SrcReg from decoding.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix 'and AL,imm8' instruction decoding
Wei Yongjun [Tue, 6 Jul 2010 08:51:09 +0000 (16:51 +0800)]
KVM: x86 emulator: fix 'and AL,imm8' instruction decoding

'and AL,imm8' should be mask as ByteOp, otherwise the dest operand
length will no correct and we may fill the full EAX when writeback.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix the comment of out instruction
Wei Yongjun [Tue, 6 Jul 2010 08:50:21 +0000 (16:50 +0800)]
KVM: x86 emulator: fix the comment of out instruction

Fix the comment of out instruction, using the same style as the
other instructions.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix 'mov sreg,rm16' instruction decoding
Wei Yongjun [Tue, 6 Jul 2010 08:49:05 +0000 (16:49 +0800)]
KVM: x86 emulator: fix 'mov sreg,rm16' instruction decoding

Memory reads for 'mov sreg,rm16' should be 16 bits only.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: Don't drop accessed bit while updating an spte
Avi Kivity [Sun, 6 Jun 2010 12:46:44 +0000 (15:46 +0300)]
KVM: MMU: Don't drop accessed bit while updating an spte

__set_spte() will happily replace an spte with the accessed bit set with
one that has the accessed bit clear.  Add a helper update_spte() which checks
for this condition and updates the page flag if needed.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: Atomically check for accessed bit when dropping an spte
Avi Kivity [Sun, 6 Jun 2010 11:48:06 +0000 (14:48 +0300)]
KVM: MMU: Atomically check for accessed bit when dropping an spte

Currently, in the window between the check for the accessed bit, and actually
dropping the spte, a vcpu can access the page through the spte and set the bit,
which will be ignored by the mmu.

Fix by using an exchange operation to atmoically fetch the spte and drop it.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: Move accessed/dirty bit checks from rmap_remove() to drop_spte()
Avi Kivity [Sun, 6 Jun 2010 11:38:12 +0000 (14:38 +0300)]
KVM: MMU: Move accessed/dirty bit checks from rmap_remove() to drop_spte()

Since we need to make the check atomic, move it to the place that will
set the new spte.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: Introduce drop_spte()
Avi Kivity [Sun, 6 Jun 2010 11:31:27 +0000 (14:31 +0300)]
KVM: MMU: Introduce drop_spte()

When we call rmap_remove(), we (almost) always immediately follow it by
an __set_spte() to a nonpresent pte.  Since we need to perform the two
operations atomically, to avoid losing the dirty and accessed bits, introduce
a helper drop_spte() and convert all call sites.

The operation is still nonatomic at this point.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: VMX: fix tlb flush with invalid root
Xiao Guangrong [Sat, 3 Jul 2010 08:02:42 +0000 (16:02 +0800)]
KVM: VMX: fix tlb flush with invalid root

Commit 341d9b535b6c simplify reload logic while entry guest mode, it
can avoid unnecessary sync-root if KVM_REQ_MMU_RELOAD and
KVM_REQ_MMU_SYNC both set.

But, it cause a issue that when we handle 'KVM_REQ_TLB_FLUSH', the
root is invalid, it is triggered during my test:

Kernel BUG at ffffffffa00212b8 [verbose debug info unavailable]
......

Fixed by directly return if the root is not ready.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Use u64 for frame data types
Joerg Roedel [Thu, 1 Jul 2010 14:00:12 +0000 (16:00 +0200)]
KVM: Use u64 for frame data types

For 32bit machines where the physical address width is
larger than the virtual address width the frame number types
in KVM may overflow. Fix this by changing them to u64.

[sfr: fix build on 32-bit ppc]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Remove unnecessary divide operations
Joerg Roedel [Thu, 1 Jul 2010 14:00:11 +0000 (16:00 +0200)]
KVM: Remove unnecessary divide operations

This patch converts unnecessary divide and modulo operations
in the KVM large page related code into logical operations.
This allows to convert gfn_t to u64 while not breaking 32
bit builds.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Fix IOMMU memslot reference warning
Sheng Yang [Thu, 1 Jul 2010 07:00:50 +0000 (15:00 +0800)]
KVM: Fix IOMMU memslot reference warning

This patch fixes the following warning.

===================================================
[ INFO: suspicious rcu_dereference_check() usage. ]
---------------------------------------------------
include/linux/kvm_host.h:259 invoked rcu_dereference_check() without
protection!

other info that might help us debug this:

rcu_scheduler_active = 1, debug_locks = 0
no locks held by qemu-system-x86/29679.

stack backtrace:
Pid: 29679, comm: qemu-system-x86 Not tainted 2.6.35-rc3+ #200
Call Trace:
 [<ffffffff810a224e>] lockdep_rcu_dereference+0xa8/0xb1
 [<ffffffffa018a06f>] kvm_iommu_unmap_memslots+0xc9/0xde [kvm]
 [<ffffffffa018a0c4>] kvm_iommu_unmap_guest+0x40/0x4e [kvm]
 [<ffffffffa018f772>] kvm_arch_destroy_vm+0x1a/0x186 [kvm]
 [<ffffffffa01800d0>] kvm_put_kvm+0x110/0x167 [kvm]
 [<ffffffffa0180ecc>] kvm_vcpu_release+0x18/0x1c [kvm]
 [<ffffffff81156f5d>] fput+0x22a/0x3a0
 [<ffffffff81152288>] filp_close+0xb4/0xcd
 [<ffffffff8106599f>] put_files_struct+0x1b7/0x36b
 [<ffffffff81065830>] ? put_files_struct+0x48/0x36b
 [<ffffffff8131ee59>] ? do_raw_spin_unlock+0x118/0x160
 [<ffffffff81065bc0>] exit_files+0x6d/0x75
 [<ffffffff81068348>] do_exit+0x47d/0xc60
 [<ffffffff8177e7b5>] ? _raw_spin_unlock_irq+0x30/0x36
 [<ffffffff81068bfa>] do_group_exit+0xcf/0x134
 [<ffffffff81080790>] get_signal_to_deliver+0x732/0x81d
 [<ffffffff81095996>] ? cpu_clock+0x4e/0x60
 [<ffffffff81002082>] do_notify_resume+0x117/0xc43
 [<ffffffff810a2fa3>] ? trace_hardirqs_on+0xd/0xf
 [<ffffffff81080d79>] ? sys_rt_sigtimedwait+0x2b5/0x3bf
 [<ffffffff8177d9f2>] ? trace_hardirqs_off_thunk+0x3a/0x3c
 [<ffffffff81003221>] ? sysret_signal+0x5/0x3d
 [<ffffffff8100343b>] int_signal+0x12/0x17

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: PPC: Make use of hash based Shadow MMU
Alexander Graf [Wed, 30 Jun 2010 13:18:46 +0000 (15:18 +0200)]
KVM: PPC: Make use of hash based Shadow MMU

We just introduced generic functions to handle shadow pages on PPC.
This patch makes the respective backends make use of them, getting
rid of a lot of duplicate code along the way.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: PPC: Add generic hpte management functions
Alexander Graf [Wed, 30 Jun 2010 13:18:45 +0000 (15:18 +0200)]
KVM: PPC: Add generic hpte management functions

Currently the shadow paging code keeps an array of entries it knows about.
Whenever the guest invalidates an entry, we loop through that entry,
trying to invalidate matching parts.

While this is a really simple implementation, it is probably the most
ineffective one possible. So instead, let's keep an array of lists around
that are indexed by a hash. This way each PTE can be added by 4 list_add,
removed by 4 list_del invocations and the search only needs to loop through
entries that share the same hash.

This patch implements said lookup and exports generic functions that both
the 32-bit and 64-bit backend can use.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: cleanup FNAME(fetch)() functions
Xiao Guangrong [Wed, 30 Jun 2010 08:05:00 +0000 (16:05 +0800)]
KVM: MMU: cleanup FNAME(fetch)() functions

Cleanup this function that we are already get the direct sp's access

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: fix direct sp's access corrupted
Xiao Guangrong [Wed, 30 Jun 2010 08:03:28 +0000 (16:03 +0800)]
KVM: MMU: fix direct sp's access corrupted

If the mapping is writable but the dirty flag is not set, we will find
the read-only direct sp and setup the mapping, then if the write #PF
occur, we will mark this mapping writable in the read-only direct sp,
now, other real read-only mapping will happily write it without #PF.

It may hurt guest's COW

Fixed by re-install the mapping when write #PF occur.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: fix conflict access permissions in direct sp
Xiao Guangrong [Wed, 30 Jun 2010 08:02:45 +0000 (16:02 +0800)]
KVM: MMU: fix conflict access permissions in direct sp

In no-direct mapping, we mark sp is 'direct' when we mapping the
guest's larger page, but its access is encoded form upper page-struct
entire not include the last mapping, it will cause access conflict.

For example, have this mapping:
        [W]
      / PDE1 -> |---|
  P[W]          |   | LPA
      \ PDE2 -> |---|
        [R]

P have two children, PDE1 and PDE2, both PDE1 and PDE2 mapping the
same lage page(LPA). The P's access is WR, PDE1's access is WR,
PDE2's access is RO(just consider read-write permissions here)

When guest access PDE1, we will create a direct sp for LPA, the sp's
access is from P, is W, then we will mark the ptes is W in this sp.

Then, guest access PDE2, we will find LPA's shadow page, is the same as
PDE's, and mark the ptes is RO.

So, if guest access PDE1, the incorrect #PF is occured.

Fixed by encode the last mapping access into direct shadow page

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: fix writable sync sp mapping
Xiao Guangrong [Wed, 30 Jun 2010 08:02:02 +0000 (16:02 +0800)]
KVM: MMU: fix writable sync sp mapping

While we sync many unsync sp at one time(in mmu_sync_children()),
we may mapping the spte writable, it's dangerous, if one unsync
sp's mapping gfn is another unsync page's gfn.

For example:

SP1.pte[0] = P
SP2.gfn's pfn = P
[SP1.pte[0] = SP2.gfn's pfn]

First, we write protected SP1 and SP2, but SP1 and SP2 are still the
unsync sp.

Then, sync SP1 first, it will detect SP1.pte[0].gfn only has one unsync-sp,
that is SP2, so it will mapping it writable, but we plan to sync SP2 soon,
at this point, the SP2->unsync is not reliable since later we sync SP2 but
SP2->gfn is already writable.

So the final result is: SP2 is the sync page but SP2.gfn is writable.

This bug will corrupt guest's page table, fixed by mark read-only mapping
if the mapped gfn has shadow pages.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: VMX: Execute WBINVD to keep data consistency with assigned devices
Sheng Yang [Wed, 30 Jun 2010 04:25:15 +0000 (12:25 +0800)]
KVM: VMX: Execute WBINVD to keep data consistency with assigned devices

Some guest device driver may leverage the "Non-Snoop" I/O, and explicitly
WBINVD or CLFLUSH to a RAM space. Since migration may occur before WBINVD or
CLFLUSH, we need to maintain data consistency either by:
1: flushing cache (wbinvd) when the guest is scheduled out if there is no
wbinvd exit, or
2: execute wbinvd on all dirty physical CPUs when guest wbinvd exits.

Signed-off-by: Yaozu (Eddie) Dong <eddie.dong@intel.com>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Document KVM specific review items
Avi Kivity [Thu, 24 Jun 2010 11:10:09 +0000 (14:10 +0300)]
KVM: Document KVM specific review items

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Simplify vcpu_enter_guest() mmu reload logic slightly
Avi Kivity [Wed, 23 Jun 2010 11:26:18 +0000 (14:26 +0300)]
KVM: Simplify vcpu_enter_guest() mmu reload logic slightly

No need to reload the mmu in between two different vcpu->requests checks.

kvm_mmu_reload() may trigger KVM_REQ_TRIPLE_FAULT, but that will be caught
during atomic guest entry later.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Search the LAPIC's for one that will accept a PIC interrupt
Chris Lalancette [Mon, 21 Jun 2010 15:29:40 +0000 (11:29 -0400)]
KVM: Search the LAPIC's for one that will accept a PIC interrupt

Older versions of 32-bit linux have a "Checking 'hlt' instruction"
test where they repeatedly call the 'hlt' instruction, and then
expect a timer interrupt to kick the CPU out of halt.  This happens
before any LAPIC or IOAPIC setup happens, which means that all of
the APIC's are in virtual wire mode at this point.  Unfortunately,
the current implementation of virtual wire mode is hardcoded to
only kick the BSP, so if a crash+kexec occurs on a different
vcpu, it will never get kicked.

This patch makes pic_unlock() do the equivalent of
kvm_irq_delivery_to_apic() for the IOAPIC code.  That is, it runs
through all of the vcpus looking for one that is in virtual wire
mode.  In the normal case where LAPICs and IOAPICs are configured,
this won't be used at all.  In the bootstrap phase of a modern
OS, before the LAPICs and IOAPICs are configured, this will have
exactly the same behavior as today; VCPU0 is always looked at
first, so it will always get out of the loop after the first
iteration.  This will only go through the loop more than once
during a kexec/kdump, in which case it will only do it a few times
until the kexec'ed kernel programs the LAPIC and IOAPIC.

Signed-off-by: Chris Lalancette <clalance@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: ia64: cleanup kvm_ia64_sync_dirty_log()
Takuya Yoshikawa [Wed, 23 Jun 2010 06:00:29 +0000 (15:00 +0900)]
KVM: ia64: cleanup kvm_ia64_sync_dirty_log()

kvm_ia64_sync_dirty_log() is a helper function for kvm_vm_ioctl_get_dirty_log()
which copies ia64's arch specific dirty bitmap to general one in memslot.
So doing sanity checks in this function is unnatural. We move these checks
outside of this and change the prototype appropriately.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: ia64: fix dirty_log_lock spin_lock section not to include get_dirty_log()
Takuya Yoshikawa [Wed, 23 Jun 2010 05:59:29 +0000 (14:59 +0900)]
KVM: ia64: fix dirty_log_lock spin_lock section not to include get_dirty_log()

kvm_get_dirty_log() calls copy_to_user(). So we need to narrow the
dirty_log_lock spin_lock section not to include this.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: PPC: Make BAT only guest segments work
Alexander Graf [Mon, 21 Jun 2010 13:24:55 +0000 (15:24 +0200)]
KVM: PPC: Make BAT only guest segments work

When a guest sets its SR entry to invalid, we may still find a
corresponding entry in a BAT. So we need to make sure we're not
faulting on invalid SR entries, but instead just claim them to be
BAT resolved.

This resolves breakage experienced when using libogc based guests.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: PPC: Use kernel hash function
Alexander Graf [Mon, 21 Jun 2010 13:25:19 +0000 (15:25 +0200)]
KVM: PPC: Use kernel hash function

The linux kernel already provides a hash function. Let's reuse that
instead of reinventing the wheel!

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: PPC: Remove obsolete kvmppc_mmu_find_pte
Alexander Graf [Mon, 21 Jun 2010 13:25:40 +0000 (15:25 +0200)]
KVM: PPC: Remove obsolete kvmppc_mmu_find_pte

Initially we had to search for pte entries to invalidate them. Since
the logic has improved since then, we can just get rid of the search
function.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Fix a race condition for usage of is_hwpoison_address()
Huang Ying [Tue, 22 Jun 2010 06:23:11 +0000 (14:23 +0800)]
KVM: Fix a race condition for usage of is_hwpoison_address()

is_hwpoison_address accesses the page table, so the caller must hold
current->mm->mmap_sem in read mode. So fix its usage in hva_to_pfn of
kvm accordingly.

Comment is_hwpoison_address to remind other users.

Reported-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86: Enable AVX for guest
Sheng Yang [Tue, 22 Jun 2010 05:49:21 +0000 (13:49 +0800)]
KVM: x86: Enable AVX for guest

Enable Intel(R) Advanced Vector Extension(AVX) for guest.

The detection of AVX feature includes OSXSAVE bit testing. When OSXSAVE bit is
not set, even if AVX is supported, the AVX instruction would result in UD as
well. So we're safe to expose AVX bits to guest directly.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Prevent internal slots from being COWed
Avi Kivity [Mon, 21 Jun 2010 07:57:45 +0000 (10:57 +0300)]
KVM: Prevent internal slots from being COWed

If a process with a memory slot is COWed, the page will change its address
(despite having an elevated reference count).  This breaks internal memory
slots which have their physical addresses loaded into vmcs registers (see
the APIC access memory slot).

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Keep slot ID in memory slot structure
Avi Kivity [Mon, 21 Jun 2010 07:56:36 +0000 (10:56 +0300)]
KVM: Keep slot ID in memory slot structure

May be used for distinguishing between internal and user slots, or for sorting
slots in size order.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Reduce atomic operations on vcpu->requests
Avi Kivity [Mon, 10 May 2010 10:08:26 +0000 (13:08 +0300)]
KVM: Reduce atomic operations on vcpu->requests

Usually the vcpu->requests bitmap is sparse, so a test_and_clear_bit() for
each request generates a large number of unneeded atomics if a bit is set.

Replace with a separate test/clear sequence.  This is safe since there is
no clear_bit() outside the vcpu thread.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Add mini-API for vcpu->requests
Avi Kivity [Mon, 10 May 2010 09:34:53 +0000 (12:34 +0300)]
KVM: Add mini-API for vcpu->requests

Makes it a little more readable and hackable.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: i8259: simplify pic_irq_request() calling sequence
Avi Kivity [Mon, 3 May 2010 14:38:06 +0000 (17:38 +0300)]
KVM: i8259: simplify pic_irq_request() calling sequence

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: i8259: reduce excessive abstraction for pic_irq_request()
Avi Kivity [Mon, 3 May 2010 14:34:34 +0000 (17:34 +0300)]
KVM: i8259: reduce excessive abstraction for pic_irq_request()

Part of the i8259 code pretends it isn't part of kvm, but we know better.
Reduce excessive abstraction, eliminating callbacks and void pointers.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Remove kernel-allocated memory regions
Avi Kivity [Mon, 21 Jun 2010 08:48:05 +0000 (11:48 +0300)]
KVM: Remove kernel-allocated memory regions

Equivalent (and better) functionality is provided by user-allocated memory
regions.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Remove memory alias support
Avi Kivity [Mon, 21 Jun 2010 08:44:20 +0000 (11:44 +0300)]
KVM: Remove memory alias support

As advertised in feature-removal-schedule.txt.  Equivalent support is provided
by overlapping memory regions.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: s390: Don't exit SIE on SIGP sense running
Christian Borntraeger [Thu, 17 Jun 2010 21:16:20 +0000 (23:16 +0200)]
KVM: s390: Don't exit SIE on SIGP sense running

Newer (guest) kernels use sigp sense running in their spinlock
implementation to check if the other cpu is running before yielding
the processor. This revealed some wrong guest settings, causing
unnecessary exits for every sigp sense running.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: s390: Fix build failure due to centralized vcpu locking patches
Christian Borntraeger [Sat, 12 Jun 2010 06:54:13 +0000 (08:54 +0200)]
KVM: s390: Fix build failure due to centralized vcpu locking patches

This patch fixes
ERROR: "__kvm_s390_vcpu_store_status" [arch/s390/kvm/kvm.ko] undefined!

triggered by
commit 3268c56840dcee78c3e928336550f4e1861504c4 (kvm.git)
Author: Avi Kivity <avi@redhat.com>
Date:   Thu May 13 12:21:46 2010 +0300
    KVM: s390: Centrally lock arch specific vcpu ioctls

Reported-by: Sachin Sant <sachinp@in.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Consolidate load/save temporary buffer allocation and freeing
Avi Kivity [Sun, 20 Jun 2010 12:54:43 +0000 (15:54 +0300)]
KVM: Consolidate load/save temporary buffer allocation and freeing

Instead of three temporary variables and three free calls, have one temporary
variable (with four names) and one free call.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Fix xsave and xcr save/restore memory leak
Avi Kivity [Sun, 20 Jun 2010 12:47:34 +0000 (15:47 +0300)]
KVM: Fix xsave and xcr save/restore memory leak

We allocate temporary kernel buffers for these structures, but never free them.

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: x86 emulator: fix group3 instruction decoding
Wei Yongjun [Thu, 17 Jun 2010 09:33:55 +0000 (17:33 +0800)]
KVM: x86 emulator: fix group3 instruction decoding

Group 3 instruction with ModRM reg field as 001 is
defined as test instruction under AMD arch, and
emulate_grp3() is ready for emulate it, so fix the
decoding.

static inline int emulate_grp3(...)
{
...
switch (c->modrm_reg) {
case 0 ... 1:   /* test */
emulate_2op_SrcV("test", c->src, c->dst, ctxt->eflags);
...
}

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: PPC: fix uninitialized variable warning in kvm_ppc_core_deliver_interrupts
Asias He [Sat, 19 Jun 2010 08:52:12 +0000 (16:52 +0800)]
KVM: PPC: fix uninitialized variable warning in kvm_ppc_core_deliver_interrupts

Fixes:
arch/powerpc/kvm/booke.c: In function 'kvmppc_core_deliver_interrupts':
arch/powerpc/kvm/booke.c:147: warning: 'msr_mask' may be used uninitialized in this function

Signed-off-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Fix typos in Documentation/kvm/mmu.txt
Jason Wang [Thu, 17 Jun 2010 08:49:22 +0000 (16:49 +0800)]
KVM: Fix typos in Documentation/kvm/mmu.txt

Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: In DM_LOWEST, only deliver interrupts to vcpus with enabled LAPIC's
Chris Lalancette [Wed, 16 Jun 2010 21:11:13 +0000 (17:11 -0400)]
KVM: x86: In DM_LOWEST, only deliver interrupts to vcpus with enabled LAPIC's

Otherwise we might try to deliver a timer interrupt to a cpu that
can't possibly handle it.

Signed-off-by: Chris Lalancette <clalance@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: Allow any LAPIC to accept PIC interrupts
Chris Lalancette [Wed, 16 Jun 2010 21:11:12 +0000 (17:11 -0400)]
KVM: x86: Allow any LAPIC to accept PIC interrupts

If the guest wants to accept timer interrupts on a CPU other
than the BSP, we need to remove this gate.

Signed-off-by: Chris Lalancette <clalance@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: Introduce a workqueue to deliver PIT timer interrupts
Chris Lalancette [Wed, 16 Jun 2010 21:11:11 +0000 (17:11 -0400)]
KVM: x86: Introduce a workqueue to deliver PIT timer interrupts

We really want to "kvm_set_irq" during the hrtimer callback,
but that is risky because that is during interrupt context.
Instead, offload the work to a workqueue, which is a bit safer
and should provide most of the same functionality.

Signed-off-by: Chris Lalancette <clalance@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86 emulator: fix pusha instruction emulation
Wei Yongjun [Tue, 15 Jun 2010 01:03:33 +0000 (09:03 +0800)]
KVM: x86 emulator: fix pusha instruction emulation

emulate pusha instruction only writeback the last
EDI register, but the other registers which need
to be writeback is ignored. This patch fixed it.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: fix -DDEBUG oops
Zachary Amsden [Mon, 14 Jun 2010 21:42:15 +0000 (11:42 -1000)]
KVM: x86: fix -DDEBUG oops

Fix a slight error with assertion in local APIC code.

Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: don't walk every parent pages while mark unsync
Xiao Guangrong [Fri, 11 Jun 2010 13:35:15 +0000 (21:35 +0800)]
KVM: MMU: don't walk every parent pages while mark unsync

While we mark the parent's unsync_child_bitmap, if the parent is already
unsynced, it no need walk it's parent, it can reduce some unnecessary
workload

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: clear unsync_child_bitmap completely
Xiao Guangrong [Fri, 11 Jun 2010 13:34:04 +0000 (21:34 +0800)]
KVM: MMU: clear unsync_child_bitmap completely

In current code, some page's unsync_child_bitmap is not cleared completely
in mmu_sync_children(), for example, if two PDPEs shard one PDT, one of
PDPE's unsync_child_bitmap is not cleared.

Currently, it not harm anything just little overload, but it's the prepare
work for the later patch

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: cleanup for __mmu_unsync_walk()
Xiao Guangrong [Fri, 11 Jun 2010 13:32:34 +0000 (21:32 +0800)]
KVM: MMU: cleanup for __mmu_unsync_walk()

Decrease sp->unsync_children after clear unsync_child_bitmap bit

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: don't mark pte notrap if it's just sync transient
Xiao Guangrong [Fri, 11 Jun 2010 13:31:38 +0000 (21:31 +0800)]
KVM: MMU: don't mark pte notrap if it's just sync transient

If the sync-sp just sync transient, don't mark its pte notrap

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: avoid double write protected in sync page path
Xiao Guangrong [Fri, 11 Jun 2010 13:30:36 +0000 (21:30 +0800)]
KVM: MMU: avoid double write protected in sync page path

The sync page is already write protected in mmu_sync_children(), don't
write protected it again

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: cleanup for dirty page judgment
Xiao Guangrong [Fri, 11 Jun 2010 13:29:42 +0000 (21:29 +0800)]
KVM: MMU: cleanup for dirty page judgment

Using wrap function to cleanup page dirty judgment

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: MMU: rename 'page' and 'shadow_page' to 'sp'
Xiao Guangrong [Fri, 11 Jun 2010 13:28:14 +0000 (21:28 +0800)]
KVM: MMU: rename 'page' and 'shadow_page' to 'sp'

Rename 'page' and 'shadow_page' to 'sp' to better fit the context

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: XSAVE/XRSTOR live migration support
Sheng Yang [Sun, 13 Jun 2010 09:29:39 +0000 (17:29 +0800)]
KVM: x86: XSAVE/XRSTOR live migration support

This patch enable save/restore of xsave state.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: PPC: fix build warning in kvm_arch_vcpu_ioctl_run
Denis Kirjanov [Fri, 11 Jun 2010 11:23:26 +0000 (11:23 +0000)]
KVM: PPC: fix build warning in kvm_arch_vcpu_ioctl_run

Fix compile warning:
  CC [M]  arch/powerpc/kvm/powerpc.o
  arch/powerpc/kvm/powerpc.c: In function 'kvm_arch_vcpu_ioctl_run':
  arch/powerpc/kvm/powerpc.c:290: warning: 'gpr' may be used uninitialized in this function
  arch/powerpc/kvm/powerpc.c:290: note: 'gpr' was declared here

Signed-off-by: Denis Kirjanov <dkirjanov@kernel.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Fix mov cr3 #GP at wrong instruction
Avi Kivity [Thu, 10 Jun 2010 14:02:16 +0000 (17:02 +0300)]
KVM: Fix mov cr3 #GP at wrong instruction

On Intel, we call skip_emulated_instruction() even if we injected a #GP,
resulting in the #GP pointing at the wrong address.

Fix by injecting the exception and skipping the instruction at the same place,
so we can do just one or the other.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Fix mov cr4 #GP at wrong instruction
Avi Kivity [Thu, 10 Jun 2010 14:02:15 +0000 (17:02 +0300)]
KVM: Fix mov cr4 #GP at wrong instruction

On Intel, we call skip_emulated_instruction() even if we injected a #GP,
resulting in the #GP pointing at the wrong address.

Fix by injecting the exception and skipping the instruction at the same place,
so we can do just one or the other.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: Fix mov cr0 #GP at wrong instruction
Avi Kivity [Thu, 10 Jun 2010 14:02:14 +0000 (17:02 +0300)]
KVM: Fix mov cr0 #GP at wrong instruction

On Intel, we call skip_emulated_instruction() even if we injected a #GP,
resulting in the #GP pointing at the wrong address.

Fix by injecting the exception and skipping the instruction at the same place,
so we can do just one or the other.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: VMX: Enable XSAVE/XRSTOR for guest
Dexuan Cui [Thu, 10 Jun 2010 03:27:12 +0000 (11:27 +0800)]
KVM: VMX: Enable XSAVE/XRSTOR for guest

This patch enable guest to use XSAVE/XRSTOR instructions.

We assume that host_xcr0 would use all possible bits that OS supported.

And we loaded xcr0 in the same way we handled fpu - do it as late as we can.

Signed-off-by: Dexuan Cui <dexuan.cui@intel.com>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: VMX: Fix incorrect rcu deref in rmode_tss_base()
Avi Kivity [Thu, 10 Jun 2010 14:21:29 +0000 (17:21 +0300)]
KVM: VMX: Fix incorrect rcu deref in rmode_tss_base()

Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Fix unused but set warnings
Andi Kleen [Thu, 10 Jun 2010 11:10:55 +0000 (13:10 +0200)]
KVM: Fix unused but set warnings

No real bugs in this one.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: Fix KVM_SET_SIGNAL_MASK with arg == NULL
Andi Kleen [Thu, 10 Jun 2010 11:10:47 +0000 (13:10 +0200)]
KVM: Fix KVM_SET_SIGNAL_MASK with arg == NULL

When the user passed in a NULL mask pass this on from the ioctl
handler.

Found by gcc 4.6's new warnings.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: delay local tlb flush
Xiao Guangrong [Tue, 8 Jun 2010 12:07:01 +0000 (20:07 +0800)]
KVM: MMU: delay local tlb flush

delay local tlb flush until enter guest moden, it can reduce vpid flush
frequency and reduce remote tlb flush IPI(if KVM_REQ_TLB_FLUSH bit is
already set, IPI is not sent)

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: use wrapper function to flush local tlb
Xiao Guangrong [Tue, 8 Jun 2010 12:05:57 +0000 (20:05 +0800)]
KVM: MMU: use wrapper function to flush local tlb

Use kvm_mmu_flush_tlb() function instead of calling
kvm_x86_ops->tlb_flush(vcpu) directly.

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: MMU: remove unnecessary remote tlb flush
Xiao Guangrong [Tue, 8 Jun 2010 12:05:05 +0000 (20:05 +0800)]
KVM: MMU: remove unnecessary remote tlb flush

This remote tlb flush is no necessary since we have synced while
sp is zapped

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: VMX: fix rcu usage warning in init_rmode()
Xiao Guangrong [Tue, 8 Jun 2010 02:15:51 +0000 (10:15 +0800)]
KVM: VMX: fix rcu usage warning in init_rmode()

fix:

[ INFO: suspicious rcu_dereference_check() usage. ]
---------------------------------------------------
include/linux/kvm_host.h:258 invoked rcu_dereference_check() without protection!

other info that might help us debug this:

rcu_scheduler_active = 1, debug_locks = 1
1 lock held by qemu-system-x86/3796:
 #0:  (&vcpu->mutex){+.+.+.}, at: [<ffffffffa0217fd8>] vcpu_load+0x1a/0x66 [kvm]

stack backtrace:
Pid: 3796, comm: qemu-system-x86 Not tainted 2.6.34 #25
Call Trace:
 [<ffffffff81070ed1>] lockdep_rcu_dereference+0x9d/0xa5
 [<ffffffffa0214fdf>] gfn_to_memslot_unaliased+0x65/0xa0 [kvm]
 [<ffffffffa0216139>] gfn_to_hva+0x22/0x4c [kvm]
 [<ffffffffa0216217>] kvm_write_guest_page+0x2a/0x7f [kvm]
 [<ffffffffa0216286>] kvm_clear_guest_page+0x1a/0x1c [kvm]
 [<ffffffffa0278239>] init_rmode+0x3b/0x180 [kvm_intel]
 [<ffffffffa02786ce>] vmx_set_cr0+0x350/0x4d3 [kvm_intel]
 [<ffffffffa02274ff>] kvm_arch_vcpu_ioctl_set_sregs+0x122/0x31a [kvm]
 [<ffffffffa021859c>] kvm_vcpu_ioctl+0x578/0xa3d [kvm]
 [<ffffffff8106624c>] ? cpu_clock+0x2d/0x40
 [<ffffffff810f7d86>] ? fget_light+0x244/0x28e
 [<ffffffff810709b9>] ? trace_hardirqs_off_caller+0x1f/0x10e
 [<ffffffff8110501b>] vfs_ioctl+0x32/0xa6
 [<ffffffff81105597>] do_vfs_ioctl+0x47f/0x4b8
 [<ffffffff813ae654>] ? sub_preempt_count+0xa3/0xb7
 [<ffffffff810f7da8>] ? fget_light+0x266/0x28e
 [<ffffffff810f7c53>] ? fget_light+0x111/0x28e
 [<ffffffff81105617>] sys_ioctl+0x47/0x6a
 [<ffffffff81002c1b>] system_call_fastpath+0x16/0x1b

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: VMX: rename vpid_sync_vcpu_all() to vpid_sync_vcpu_single()
Gui Jianfeng [Mon, 7 Jun 2010 02:33:27 +0000 (10:33 +0800)]
KVM: VMX: rename vpid_sync_vcpu_all() to vpid_sync_vcpu_single()

The name "pid_sync_vcpu_all" isn't appropriate since it just affect
a single vpid, so rename it to vpid_sync_vcpu_single().

Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
14 years agoKVM: VMX: Add all-context INVVPID type support
Gui Jianfeng [Mon, 7 Jun 2010 02:32:29 +0000 (10:32 +0800)]
KVM: VMX: Add all-context INVVPID type support

Add all-context INVVPID type support.

Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>