We're in a hot path. We can't use kmalloc() because
it might impact performance. So, we just stick the buffer that
we need into the kvm_vcpu_arch structure. This is used very
often, so it is not really a waste.
We also have to move the buffer structure's definition to the
arch-specific x86 kvm header.
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
[sheng: fix KVM_GET_LAPIC using wrong size]
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
On my machine with gcc 3.4, kvm uses ~2k of stack in a few
select functions. This is mostly because gcc fails to
notice that the different case: statements could have their
stack usage combined. It overflows very nicely if interrupts
happen during one of these large uses.
This patch uses two methods for reducing stack usage.
1. dynamically allocate large objects instead of putting
on the stack.
2. Use a union{} member for all of the case variables. This
tricks gcc into combining them all into a single stack
allocation. (There's also a comment on this)
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Based on a patch from: Amit Shah <amit.shah@qumranet.com>
This patch adds support for handling PCI devices that are assigned to
the guest.
The device to be assigned to the guest is registered in the host kernel
and interrupt delivery is handled. If a device is already assigned, or
the device driver for it is still loaded on the host, the device
assignment is failed by conveying a -EBUSY reply to the userspace.
Devices that share their interrupt line are not supported at the moment.
By itself, this patch will not make devices work within the guest.
The VT-d extension is required to enable the device to perform DMA.
Another alternative is PVDMA.
Signed-off-by: Amit Shah <amit.shah@qumranet.com>
Signed-off-by: Ben-Ami Yassour <benami@il.ibm.com>
Signed-off-by: Weidong Han <weidong.han@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The PIT injection logic is problematic under the following cases:
1) If there is a higher priority vector to be delivered by the time
kvm_pit_timer_intr_post is invoked ps->inject_pending won't be set.
This opens the possibility for missing many PIT event injections (say if
guest executes hlt at this point).
2) ps->inject_pending is racy with more than two vcpus. Since there's no locking
around read/dec of pt->pending, two vcpu's can inject two interrupts for a single
pt->pending count.
Fix 1 by using an irq ack notifier: only reinject when the previous irq
has been acked. Fix 2 with appropriate locking around manipulation of
pending count and irq_ack by the injection / ack paths.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Based on a patch from: Ben-Ami Yassour <benami@il.ibm.com>
which was based on a patch from: Amit Shah <amit.shah@qumranet.com>
Notify IRQ acking on PIC/APIC emulation. The previous patch missed two things:
- Edge triggered interrupts on IOAPIC
- PIC reset with IRR/ISR set should be equivalent to ack (LAPIC probably
needs something similar).
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
CC: Amit Shah <amit.shah@qumranet.com>
CC: Ben-Ami Yassour <benami@il.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This can be used by kvm subsystems that are interested in when
interrupts are acked, for example time drift compensation.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Netware writes to DEBUGCTL and reads from the DEBUGCTL and LAST*IP MSRs
without further checks and is really confused to receive a #GP during that.
To make it happy we should just make them stubs, which is exactly what SVM
already does.
Writes to DEBUGCTL that are vendor-specific are resembled to behave as if the
virtual CPU does not know them.
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Usually HOST_RSP retains its value across guest entries. Take advantage
of this and avoid a vmwrite() when this is so.
Signed-off-by: Avi Kivity <avi@qumranet.com>
As we execute real mode guests in VM86 mode, exception have to be
reinjected appropriately when the guest triggered them. For this purpose
the patch adopts the real-mode injection pattern used in vmx_inject_irq
to vmx_queue_exception, additionally taking care that the IP is set
correctly for #BP exceptions. Furthermore it extends
handle_rmode_exception to reinject all those exceptions that can be
raised in real mode.
This fixes the execution of himem.exe from FreeDOS and also makes its
debug.com work properly.
Note that guest debugging in real mode is broken now. This has to be
fixed by the scheduled debugging infrastructure rework (will be done
once base patches for QEMU have been accepted).
Signed-off-by: Jan Kiszka <jan.kiszka@web.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Since checking for vcpu->arch.rmode.active is already done whenever we
call handle_rmode_exception(), checking it inside the function is redundant.
Signed-off-by: Mohammed Gamal <m.gamal005@gmail.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Instead of looking at failed injections in the vm entry path, move
processing to the exit path in vmx_complete_interrupts(). This simplifes
the logic and removes any state that is hidden in vmx registers.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Similar to the exception queue, this hold interrupts that have been
accepted by the virtual processor core but not yet injected.
Not yet used.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The vmx code assumes that IDT-Vectoring can only be set when an exception
is injected due to the exception in question. That's not true, however:
if the exception is injected correctly, and later another exception occurs
but its delivery is blocked due to a fault, then we will incorrectly assume
the first exception was not delivered.
Fix by unconditionally dequeuing the pending exception, and requeuing it
(or the second exception) if we see it in the IDT-Vectoring field.
Signed-off-by: Avi Kivity <avi@qumranet.com>
If we're emulating an instruction, either it will succeed, in which case
any previously queued exception will be spurious, or we will requeue the
same exception.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Instead of processing nmi injection failure in the vm entry path, move
it to the vm exit path (vm_complete_interrupts()). This separates nmi
injection from nmi post-processing, and moves the nmi state from the VT
state into vcpu state (new variable nmi_injected specifying an injection
in progress).
Signed-off-by: Avi Kivity <avi@qumranet.com>
Currently most interrupt exit processing is handled on the entry path,
which is confusing. Move the NMI IRET fault processing to a new function,
vmx_complete_interrupts(), which is called on the vmexit path.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The twisty maze of conditionals can be reduced.
[joerg: fix tlb flushing]
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This function injects an interrupt into the guest given the kvm struct,
the (guest) irq number and the interrupt level.
Signed-off-by: Amit Shah <amit.shah@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
As suggested by Avi, introduce accessors to read/write guest registers.
This simplifies the ->cache_regs/->decache_regs interface, and improves
register caching which is important for VMX, where the cost of
vmcs_read/vmcs_write is significant.
[avi: fix warnings]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
MSR_IA32_FEATURE_LOCKED is just a bit in fact, which shouldn't be prefixed with
MSR_. So is MSR_IA32_FEATURE_VMXON_ENABLED.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
As well as discard fake accessed bit and dirty bit of EPT.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Accesses to CR4 are intercepted even with Nested Paging enabled. But the code
does not check if the guest wants to do a global TLB flush. So this flush gets
lost. This patch adds the check and the flush to svm_set_cr4.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch introduces a guest TLB flush on every NPF exit in KVM. This fixes
random segfaults and #UD exceptions in the guest seen under some workloads
(e.g. long running compile workloads or tbench). A kernbench run with and
without that fix showed that it has a slowdown lower than 0.5%
Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
They are hardware specific MSRs, and we would use them in virtualization
feature detection later.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The shadow code assigns a pte directly in one place, which is nonatomic on
i386 can can cause random memory references. Fix by using an atomic setter.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Synchronize changes to host virtual addresses which are part of
a KVM memory slot to the KVM shadow mmu. This allows pte operations
like swapping, page migration, and madvise() to transparently work
with KVM.
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This allows reading memslots with only the mmu_lock hold for mmu
notifiers that runs in atomic context and with mmu_lock held.
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This allows the mmu notifier code to run unalias_gfn with only the
mmu_lock held. Only alias writes need the mmu_lock held. Readers will
either take the slots_lock in read mode or the mmu_lock.
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages.
There are secondary MMUs (with secondary sptes and secondary tlbs) too.
sptes in the kvm case are shadow pagetables, but when I say spte in
mmu-notifier context, I mean "secondary pte". In GRU case there's no
actual secondary pte and there's only a secondary tlb because the GRU
secondary MMU has no knowledge about sptes and every secondary tlb miss
event in the MMU always generates a page fault that has to be resolved by
the CPU (this is not the case of KVM where the a secondary tlb miss will
walk sptes in hardware and it will refill the secondary tlb transparently
to software if the corresponding spte is present). The same way
zap_page_range has to invalidate the pte before freeing the page, the spte
(and secondary tlb) must also be invalidated before any page is freed and
reused.
Currently we take a page_count pin on every page mapped by sptes, but that
means the pages can't be swapped whenever they're mapped by any spte
because they're part of the guest working set. Furthermore a spte unmap
event can immediately lead to a page to be freed when the pin is released
(so requiring the same complex and relatively slow tlb_gather smp safe
logic we have in zap_page_range and that can be avoided completely if the
spte unmap event doesn't require an unpin of the page previously mapped in
the secondary MMU).
The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know
when the VM is swapping or freeing or doing anything on the primary MMU so
that the secondary MMU code can drop sptes before the pages are freed,
avoiding all page pinning and allowing 100% reliable swapping of guest
physical address space. Furthermore it avoids the code that teardown the
mappings of the secondary MMU, to implement a logic like tlb_gather in
zap_page_range that would require many IPI to flush other cpu tlbs, for
each fixed number of spte unmapped.
To make an example: if what happens on the primary MMU is a protection
downgrade (from writeable to wrprotect) the secondary MMU mappings will be
invalidated, and the next secondary-mmu-page-fault will call
get_user_pages and trigger a do_wp_page through get_user_pages if it
called get_user_pages with write=1, and it'll re-establishing an updated
spte or secondary-tlb-mapping on the copied page. Or it will setup a
readonly spte or readonly tlb mapping if it's a guest-read, if it calls
get_user_pages with write=0. This is just an example.
This allows to map any page pointed by any pte (and in turn visible in the
primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an
full MMU with both sptes and secondary-tlb like the shadow-pagetable layer
with kvm), or a remote DMA in software like XPMEM (hence needing of
schedule in XPMEM code to send the invalidate to the remote node, while no
need to schedule in kvm/gru as it's an immediate event like invalidating
primary-mmu pte).
At least for KVM without this patch it's impossible to swap guests
reliably. And having this feature and removing the page pin allows
several other optimizations that simplify life considerably.
Dependencies:
1) mm_take_all_locks() to register the mmu notifier when the whole VM
isn't doing anything with "mm". This allows mmu notifier users to keep
track if the VM is in the middle of the invalidate_range_begin/end
critical section with an atomic counter incraese in range_begin and
decreased in range_end. No secondary MMU page fault is allowed to map
any spte or secondary tlb reference, while the VM is in the middle of
range_begin/end as any page returned by get_user_pages in that critical
section could later immediately be freed without any further
->invalidate_page notification (invalidate_range_begin/end works on
ranges and ->invalidate_page isn't called immediately before freeing
the page). To stop all page freeing and pagetable overwrites the
mmap_sem must be taken in write mode and all other anon_vma/i_mmap
locks must be taken too.
2) It'd be a waste to add branches in the VM if nobody could possibly
run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if
CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of
mmu notifiers, but this already allows to compile a KVM external module
against a kernel with mmu notifiers enabled and from the next pull from
kvm.git we'll start using them. And GRU/XPMEM will also be able to
continue the development by enabling KVM=m in their config, until they
submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can
also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n).
This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM
are all =n.
The mmu_notifier_register call can fail because mm_take_all_locks may be
interrupted by a signal and return -EINTR. Because mmu_notifier_reigster
is used when a driver startup, a failure can be gracefully handled. Here
an example of the change applied to kvm to register the mmu notifiers.
Usually when a driver startups other allocations are required anyway and
-ENOMEM failure paths exists already.
struct kvm *kvm_arch_create_vm(void)
{
struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
+ int err;
if (!kvm)
return ERR_PTR(-ENOMEM);
INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
+ kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops;
+ err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm);
+ if (err) {
+ kfree(kvm);
+ return ERR_PTR(err);
+ }
+
return kvm;
}
mmu_notifier_unregister returns void and it's reliable.
The patch also adds a few needed but missing includes that would prevent
kernel to compile after these changes on non-x86 archs (x86 didn't need
them by luck).
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix mm/filemap_xip.c build]
[akpm@linux-foundation.org: fix mm/mmu_notifier.c build]
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Kanoj Sarcar <kanojsarcar@yahoo.com>
Cc: Roland Dreier <rdreier@cisco.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Avi Kivity <avi@qumranet.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Chris Wright <chrisw@redhat.com>
Cc: Marcelo Tosatti <marcelo@kvack.org>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Izik Eidus <izike@qumranet.com>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When an event (such as an interrupt) is injected, and the stack is
shadowed (and therefore write protected), the guest will exit. The
current code will see that the stack is shadowed and emulate a few
instructions, each time postponing the injection. Eventually the
injection may succeed, but at that time the guest may be unwilling
to accept the interrupt (for example, the TPR may have changed).
This occurs every once in a while during a Windows 2008 boot.
Fix by unshadowing the fault address if the fault was due to an event
injection.
Signed-off-by: Avi Kivity <avi@qumranet.com>
There is no guarantee that the old TSS descriptor in the GDT contains
the proper base address. This is the case for Windows installation's
reboot-via-triplefault.
Use guest registers instead. Also translate the address properly.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The segment base is always a linear address, so translate before
accessing guest memory.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
If NPT is enabled after loading both KVM modules on AMD and it should be
disabled, both KVM modules must be reloaded. If only the architecture module is
reloaded the behavior is undefined. With this patch it is possible to disable
NPT only by reloading the kvm_amd module.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The direct mapped shadow code (used for real mode and two dimensional paging)
sets upper-level ptes using direct assignment rather than calling
set_shadow_pte(). A nonpae host will split this into two writes, which opens
up a race if another vcpu accesses the same memory area.
Fix by calling set_shadow_pte() instead of assigning directly.
Noticed by Izik Eidus.
Signed-off-by: Avi Kivity <avi@qumranet.com>
If the guest issues a clflush in a mmio address, the instruction
can trap into the hypervisor. Currently, we do not decode clflush
properly, causing the guest to hang. This patch fixes this emulating
clflush (opcode 0f ae).
Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Harden kvm_mmu_zap_page() against invalid root pages that
had been shadowed from memslots that are gone.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Flush the shadow mmu before removing regions to avoid stale entries.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Dave Hansen
Ben-Ami Yassour
Marcelo Tosatti
Avi Kivity
Alexander Graf
Jan Kiszka
Mohammed Gamal
Amit Shah
Sheng Yang
Ingo Molnar
Joerg Roedel
Andrea Arcangeli
Glauber Costa