Sheng Yang [Wed, 2 Jan 2008 06:49:22 +0000 (14:49 +0800)]
KVM: x86 emulator: Only allow VMCALL/VMMCALL trapped by #UD
When executing a test program called "crashme", we found the KVM guest cannot
survive more than ten seconds, then encounterd kernel panic. The basic concept
of "crashme" is generating random assembly code and trying to execute it.
After some fixes on emulator insn validity judgment, we found it's hard to
get the current emulator handle the invalid instructions correctly, for the
#UD trap for hypercall patching caused troubles. The problem is, if the opcode
itself was OK, but combination of opcode and modrm_reg was invalid, and one
operand of the opcode was memory (SrcMem or DstMem), the emulator will fetch
the memory operand first rather than checking the validity, and may encounter
an error there. For example, ".byte 0xfe, 0x34, 0xcd" has this problem.
In the patch, we simply check that if the invalid opcode wasn't vmcall/vmmcall,
then return from emulate_instruction() and inject a #UD to guest. With the
patch, the guest had been running for more than 12 hours.
Signed-off-by: Feng (Eric) Liu <eric.e.liu@intel.com> Signed-off-by: Sheng Yang <sheng.yang@intel.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Mon, 31 Dec 2007 13:27:49 +0000 (15:27 +0200)]
KVM: MMU: Move kvm_free_some_pages() into critical section
If some other cpu steals mmu pages between our check and an attempt to
allocate, we can run out of mmu pages. Fix by moving the check into the
same critical section as the allocation.
Avi Kivity [Sun, 30 Dec 2007 10:29:05 +0000 (12:29 +0200)]
KVM: MMU: Avoid calling gfn_to_page() in mmu_set_spte()
Since gfn_to_page() is a sleeping function, and we want to make the core mmu
spinlocked, we need to pass the page from the walker context (which can sleep)
to the shadow context (which cannot).
Marcelo Tosatti [Fri, 21 Dec 2007 00:18:22 +0000 (19:18 -0500)]
KVM: MMU: Concurrent guest walkers
Do not hold kvm->lock mutex across the entire pagefault code,
only acquire it in places where it is necessary, such as mmu
hash list, active list, rmap and parent pte handling.
Allow concurrent guest walkers by switching walk_addr() to use
mmap_sem in read-mode.
And get rid of the lockless __gfn_to_page.
[avi: move kvm_mmu_pte_write() locking inside the function]
[avi: add locking for real mode]
[avi: fix cmpxchg locking]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Thu, 25 Oct 2007 14:52:32 +0000 (16:52 +0200)]
KVM: Accelerated apic support
This adds a mechanism for exposing the virtual apic tpr to the guest, and a
protocol for letting the guest update the tpr without causing a vmexit if
conditions allow (e.g. there is no interrupt pending with a higher priority
than the new tpr).
Avi Kivity [Mon, 22 Oct 2007 14:50:39 +0000 (16:50 +0200)]
KVM: local APIC TPR access reporting facility
Add a facility to report on accesses to the local apic tpr even if the
local apic is emulated in the kernel. This is basically a hack that
allows userspace to patch Windows which tends to bang on the tpr a lot.
Ryan Harper [Thu, 13 Dec 2007 16:21:10 +0000 (10:21 -0600)]
KVM: VMX: Add printk_ratelimit in vmx_intr_assist
Add printk_ratelimit check in front of printk. This prevents spamming
of the message during 32-bit ubuntu 6.06server install. Previously, it
would hang during the partition formatting stage.
Signed-off-by: Ryan Harper <ryanh@us.ibm.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Joerg Roedel [Tue, 11 Dec 2007 14:36:57 +0000 (15:36 +0100)]
KVM: SVM: support writing 0 to K8 performance counter control registers
This lets SVM ignore writes of the value 0 to the performance counter control
registers. Thus enabling them will still fail in the guest, but a write of 0
which keeps them disabled is accepted. This is required to boot Windows
Vista 64bit.
[avi: avoid fall-thru in switch statement]
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com> Signed-off-by: Markus Rechberger <markus.rechberger@amd.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Sun, 9 Dec 2007 16:43:00 +0000 (18:43 +0200)]
KVM: MMU: Use mmu_set_spte() for real-mode shadows
In addition to removing some duplicated code, this also handles the unlikely
case of real-mode code updating a guest page table. This can happen when
one vcpu (in real mode) touches a second vcpu's (in protected mode) page
tables, or if a vcpu switches to real mode, touches page tables, and switches
back.
Avi Kivity [Sun, 9 Dec 2007 15:40:31 +0000 (17:40 +0200)]
KVM: MMU: Move set_pte() into guest paging mode independent code
As set_pte() no longer references either a gpte or the guest walker, we can
move it out of paging mode dependent code (which compiles twice and is
generally nasty).
Avi Kivity [Sun, 9 Dec 2007 15:00:02 +0000 (17:00 +0200)]
KVM: MMU: Fix inherited permissions for emulated guest pte updates
When we emulate a guest pte write, we fail to apply the correct inherited
permissions from the parent ptes. Now that we store inherited permissions
in the shadow page, we can use that to update the pte permissions correctly.
Avi Kivity [Sun, 9 Dec 2007 14:37:36 +0000 (16:37 +0200)]
KVM: MMU: Set nx bit correctly on shadow ptes
While the page table walker correctly generates a guest page fault
if a guest tries to execute a non-executable page, the shadow code does
not mark it non-executable. This means that if a guest accesses an nx
page first with a read access, then subsequent code fetch accesses will
succeed.
Avi Kivity [Sun, 9 Dec 2007 14:15:46 +0000 (16:15 +0200)]
KVM: MMU: Simplify calculation of pte access
The nx bit is awkwardly placed in the 63rd bit position; furthermore it
has a reversed meaning compared to the other bits, which means we can't use
a bitwise and to calculate compounded access masks.
So, we simplify things by creating a new 3-bit exec/write/user access word,
and doing all calculations in that.
Marcelo Tosatti [Fri, 7 Dec 2007 12:56:58 +0000 (07:56 -0500)]
KVM: MMU: Use cmpxchg for pte updates on walk_addr()
In preparation for multi-threaded guest pte walking, use cmpxchg()
when updating guest pte's. This guarantees that the assignment of the
dirty bit can't be lost if two CPU's are faulting the same address
simultaneously.
[avi: fix kunmap_atomic() parameters]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Thu, 6 Dec 2007 16:14:14 +0000 (18:14 +0200)]
KVM: x86 emulator: Fix stack instructions on 64-bit mode
Stack instructions are always 64-bit on 64-bit mode; many of the
emulated stack instructions did not take that into account. Fix by
adding a 'Stack' bitflag and setting the operand size appropriately
during the decode stage (except for 'push r/m', which is in a group
with a few other instructions, so it gets its own treatment).
Joerg Roedel [Thu, 6 Dec 2007 14:46:52 +0000 (15:46 +0100)]
KVM: SVM: Emulate read/write access to cr8
This patch adds code to emulate the access to the cr8 register to the x86
instruction emulator in kvm. This is needed on svm, where there is no
hardware decode for control register access.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com> Signed-off-by: Markus Rechberger <markus.rechberger@amd.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Thu, 6 Dec 2007 14:32:45 +0000 (16:32 +0200)]
KVM: VMX: Avoid exit when setting cr8 if the local apic is in the kernel
With apic in userspace, we must exit to userspace after a cr8 write in order
to update the tpr. But if the apic is in the kernel, the exit is unnecessary.
Avi Kivity [Sun, 25 Nov 2007 11:41:11 +0000 (13:41 +0200)]
KVM: Generalize exception injection mechanism
Instead of each subarch doing its own thing, add an API for queuing an
injection, and manage failed exception injection centerally (i.e., if
an inject failed due to a shadow page fault, we need to requeue it).
KVM: SVM: Remove KVM specific defines for MSR_EFER
This patch removes the KVM specific defines for MSR_EFER that were being used
in the svm support file and migrates all references to use instead the ones
from the kernel headers that are used everywhere else and that have the same
values.
Signed-off-by: Carlo Marcelo Arenas Belon <carenas@sajinet.com.pe> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Sun, 2 Dec 2007 08:50:06 +0000 (10:50 +0200)]
KVM: Export include/linux/kvm.h only if $ARCH actually supports KVM
Currently, make headers_check barfs due to <asm/kvm.h>, which <linux/kvm.h>
includes, not existing. Rather than add a zillion <asm/kvm.h>s, export kvm.h
only if the arch actually supports it.
Avi Kivity [Tue, 27 Nov 2007 17:30:56 +0000 (19:30 +0200)]
KVM: x86 emulator: unify four switch statements into two
Unify the special instruction switch with the regular instruction switch,
and the two byte special instruction switch with the regular two byte
instruction switch. That makes it much easier to find an instruction or
the place an instruction needs to be added in.
Jerone Young [Mon, 26 Nov 2007 14:33:53 +0000 (08:33 -0600)]
KVM: Add ifdef in irqchip struct for x86 only structures
This patch fixes a small issue where sturctures:
kvm_pic_state
kvm_ioapic_state
are defined inside x86 specific code and may or may not
be defined in anyway for other architectures. The problem
caused is one cannot compile userspace apps (ex. libkvm)
for other archs since a size cannot be determined for these
structures.
Signed-off-by: Jerone Young <jyoung5@us.ibm.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
KVM: x86 emulator: Make a distinction between repeat prefixes F3 and F2
cmps and scas instructions accept repeat prefixes F3 and F2. So in
order to emulate those prefixed instructions we need to be able to know
if prefixes are REP/REPE/REPZ or REPNE/REPNZ. Currently kvm doesn't make
this distinction. This patch introduces this distinction.
Signed-off-by: Guillaume Thouvenin <guillaume.thouvenin@ext.bull.net> Signed-off-by: Avi Kivity <avi@qumranet.com>
Dan Kenigsberg [Wed, 21 Nov 2007 15:10:04 +0000 (17:10 +0200)]
KVM: Enhance guest cpuid management
The current cpuid management suffers from several problems, which inhibit
passing through the host feature set to the guest:
- No way to tell which features the host supports
While some features can be supported with no changes to kvm, others
need explicit support. That means kvm needs to vet the feature set
before it is passed to the guest.
- No support for indexed or stateful cpuid entries
Some cpuid entries depend on ecx as well as on eax, or on internal
state in the processor (running cpuid multiple times with the same
input returns different output). The current cpuid machinery only
supports keying on eax.
- No support for save/restore/migrate
The internal state above needs to be exposed to userspace so it can
be saved or migrated.
This patch adds extended cpuid support by means of three new ioctls:
- KVM_GET_SUPPORTED_CPUID: get all cpuid entries the host (and kvm)
supports
- KVM_SET_CPUID2: sets the vcpu's cpuid table
- KVM_GET_CPUID2: gets the vcpu's cpuid table, including hidden state
[avi: fix original KVM_SET_CPUID not removing nx on non-nx hosts as it did
before]
Signed-off-by: Dan Kenigsberg <danken@qumranet.com> Signed-off-by: Avi Kivity <avi@qumranet.com>
Avi Kivity [Wed, 21 Nov 2007 13:28:32 +0000 (15:28 +0200)]
KVM: MMU: Rename variables of type 'struct kvm_mmu_page *'
These are traditionally named 'page', but even more traditionally, that name
is reserved for variables that point to a 'struct page'. Rename them to 'sp'
(for "shadow page").
Avi Kivity [Wed, 21 Nov 2007 12:44:45 +0000 (14:44 +0200)]
KVM: MMU: Introduce gfn_to_gpa()
Converting a frame number to an address is tricky since the data type changes
size. Introduce a function to do it. This fixes an actual bug when
accessing guest ptes.
Avi Kivity [Wed, 21 Nov 2007 00:06:21 +0000 (02:06 +0200)]
KVM: MMU: Avoid unnecessary remote tlb flushes when guest updates a pte
If all we're doing is increasing permissions on a pte (typical for demand
paging), then there's not need to flush remote tlbs. Worst case they'll
get a spurious page fault.
Avi Kivity [Tue, 20 Nov 2007 19:39:54 +0000 (21:39 +0200)]
KVM: MMU: Implement guest page fault bypass for nonpae
I spent an hour worrying why I see so many guest page faults on FC6 i386.
Turns out bypass wasn't implemented for nonpae. Implement it so it doesn't
happen again.
projects / linux-2.6 / log