Some guests (Solaris) do not set up all four pdptrs, but leave some invalid.
kvm incorrectly treated these as valid page directories, pinning the
wrong pages and causing general confusion.
Fix by checking the valid bit of a pae pdpte. This closes sourceforge bug
1698922.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Solaris panics if it sees a cpu with no fpu, and it seems to rely on this
bit. Closes sourceforge bug 1698920.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The expression
sp - 6 < sp
where sp is a u16 is undefined in C since 'sp - 6' is promoted to int,
and signed overflow is undefined in C. gcc 4.2 actually warns about it.
Replace with a simpler test.
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch enables the virtualization of the last branch record MSRs on
SVM if this feature is available in hardware. It also introduces a small
and simple check feature for specific SVM extensions.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
With this, we can specify that accesses to one physical memory range will
be remapped to another. This is useful for the vga window at 0xa0000 which
is used as a movable window into the (much larger) framebuffer.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Mapping a guest page to a host page is a common operation. Currently,
one has first to find the memory slot where the page belongs (gfn_to_memslot),
then locate the page itself (gfn_to_page()).
This is clumsy, and also won't work well with memory aliases. So simplify
gfn_to_page() not to require memory slot translation first, and instead do it
internally.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Functions that play around with the physical memory map
need a way to clear mappings to possibly nonexistent or
invalid memory. Both the mmu cache and the processor tlb
are cleared.
Signed-off-by: Dor Laor <dor.laor@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
On x86, bit operations operate on a string of bits that can reside in
multiple words. For example, 'btsl %eax, (blah)' will touch the word
at blah+4 if %eax is between 32 and 63.
The x86 emulator compensates for that by advancing the operand address
by (bit offset / BITS_PER_LONG) and truncating the bit offset to the
range (0..BITS_PER_LONG-1). This has a side effect of forcing the operand
size to 8 bytes on 64-bit hosts.
Now, a 32-bit guest goes and fork()s a process. It write protects a stack
page at 0xbffff000 using the 'btr' instruction, at offset 0xffc in the page
table, with bit offset 1 (for the write permission bit).
The emulator now forces the operand size to 8 bytes as previously described,
and an innocent page table update turns into a cross-page-boundary write,
which is assumed by the mmu code not to be a page table, so it doesn't
actually clear the corresponding shadow page table entry. The guest and
host permissions are out of sync and guest memory is corrupted soon
afterwards, leading to guest failure.
Fix by not using BITS_PER_LONG as the word size; instead use the actual
operand size, so we get a 32-bit write in that case.
Note we still have to teach the mmu to handle cross-page-boundary writes
to guest page table; but for now this allows Damn Small Linux 0.4 (2.4.20)
to boot.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Remove unused function
CC drivers/kvm/svm.o
drivers/kvm/svm.c:207: warning: ‘inject_db’ defined but not used
Signed-off-by: Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
When a vcpu is migrated from one cpu to another, its timestamp counter
may lose its monotonic property if the host has unsynced timestamp counters.
This can confuse the guest, sometimes to the point of refusing to boot.
As the rdtsc instruction is rather fast on AMD processors (7-10 cycles),
we can simply record the last host tsc when we drop the cpu, and adjust
the vcpu tsc offset when we detect that we've migrated to a different cpu.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The kvm mmu keeps a shadow page for hugepage pdes; if several such pdes map
the same physical address, they share the same shadow page. This is a fairly
common case (kernel mappings on i386 nonpae Linux, for example).
However, if the two pdes map the same memory but with different permissions, kvm
will happily use the cached shadow page. If the access through the more
permissive pde will occur after the access to the strict pde, an endless pagefault
loop will be generated and the guest will make no progress.
Fix by making the access permissions part of the cache lookup key.
The fix allows Xen pae to boot on kvm and run guest domains.
Thanks to Jeremy Fitzhardinge for reporting the bug and testing the fix.
Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch forbids the guest to execute monitor/mwait instructions on
SVM. This is necessary because the guest can execute these instructions
if they are available even if the kvm cpuid doesn't report its
existence.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Some older (~2.6.7) kernels write MCG_STATUS register during kernel
boot (mce_clear_all() function, called from mce_init()). It's not
currently handled by kvm and will cause it to inject a GPF.
Following patch adds a "nop" handler for this.
Signed-off-by: Sergey Kiselev <sergey.kiselev@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
As usual, we need to mangle segment registers when emulating real mode
as vm86 has specific constraints. We special case the reset segment base,
and set the "access rights" (or descriptor flags) to vm86 comaptible values.
This fixes reboot on vmx.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The SET_SREGS ioctl modifies both cr0.pe (real mode/protected mode) and
guest segment registers. Since segment handling is modified by the mode on
Intel procesors, update the segment registers after the mode switch has taken
place.
Signed-off-by: Avi Kivity <avi@qumranet.com>
set_cr0_no_modeswitch() was a hack to avoid corrupting segment registers.
As we now cache the protected mode values on entry to real mode, this
isn't an issue anymore, and it interferes with reboot (which usually _is_
a modeswitch).
Signed-off-by: Avi Kivity <avi@qumranet.com>
The reset state has cs.selector == 0xf000 and cs.base == 0xffff0000,
which aren't compatible with vm86 mode, which is used for real mode
virtualization.
When we create a vcpu, we set cs.base to 0xf0000, but if we get there by
way of a reset, the values are inconsistent and vmx refuses to enter
guest mode.
Workaround by detecting the state and munging it appropriately.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The initial, noncaching, version of the kvm mmu flushed the all nonglobal
shadow page table translations (much like a native tlb flush). The new
implementation flushes translations only when they change, rendering global
pte tracking superfluous.
This removes the unused tracking mechanism and storage space.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The current string pio interface communicates using guest virtual addresses,
relying on userspace to translate addresses and to check permissions. This
interface cannot fully support guest smp, as the check needs to take into
account two pages at one in case an unaligned string transfer straddles a
page boundary.
Change the interface not to communicate guest addresses at all; instead use
a buffer page (mmaped by userspace) and do transfers there. The kernel
manages the virtual to physical translation and can perform the checks
atomically by taking the appropriate locks.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Some ioctls ignore their arguments. By requiring them to be zero now,
we allow a nonzero value to have some special meaning in the future.
Signed-off-by: Avi Kivity <avi@qumranet.com>
This allows us to store offsets in the kernel/user kvm_run area, and be
sure that userspace has them mapped. As offsets can be outside the
kvm_run struct, userspace has no way of knowing how much to mmap.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Allow a special signal mask to be used while executing in guest mode. This
allows signals to be used to interrupt a vcpu without requiring signal
delivery to a userspace handler, which is quite expensive. Userspace still
receives -EINTR and can get the signal via sigwait().
Signed-off-by: Avi Kivity <avi@qumranet.com>
This is redundant, as we also return -EINTR from the ioctl, but it
allows us to examine the exit_reason field on resume without seeing
old data.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Currently, userspace is told about the nature of the last exit from the
guest using two fields, exit_type and exit_reason, where exit_type has
just two enumerations (and no need for more). So fold exit_type into
exit_reason, reducing the complexity of determining what really happened.
Signed-off-by: Avi Kivity <avi@qumranet.com>
KVM used to handle cpuid by letting userspace decide what values to
return to the guest. We now handle cpuid completely in the kernel. We
still let userspace decide which values the guest will see by having
userspace set up the value table beforehand (this is necessary to allow
management software to set the cpu features to the least common denominator,
so that live migration can work).
The motivation for the change is that kvm kernel code can be impacted by
cpuid features, for example the x86 emulator.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Currently when passing the a PIO emulation request to userspace, we
rely on userspace updating %rax (on 'in' instructions) and %rsi/%rdi/%rcx
(on string instructions). This (a) requires two extra ioctls for getting
and setting the registers and (b) is unfriendly to non-x86 archs, when
they get kvm ports.
So fix by doing the register fixups in the kernel and passing to userspace
only an abstract description of the PIO to be done.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Instead of passing a 'struct kvm_run' back and forth between the kernel and
userspace, allocate a page and allow the user to mmap() it. This reduces
needless copying and makes the interface expandable by providing lots of
free space.
Signed-off-by: Avi Kivity <avi@qumranet.com>
When auditing a 32-bit guest on a 64-bit host, sign extension of the page
table directory pointer table index caused bogus addresses to be shown on
audit errors.
Fix by declaring the index unsigned.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Instead of twiddling the rip registers directly, use the
skip_emulated_instruction() function to do that for us.
Signed-off-by: Dor Laor <dor.laor@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The hypercall code mixes up the ->cache_regs() and ->decache_regs()
callbacks, resulting in guest register corruption.
Signed-off-by: Dor Laor <dor.laor@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Nonpae guest pdes are shadowed by two pae ptes, so we double the offset
twice: once to account for the pte size difference, and once because we
need to shadow pdes for a single guest pde.
But when writing to the upper guest pde we also need to truncate the
lower bits, otherwise the multiply shifts these bits into the pde index
and causes an access to the wrong shadow pde. If we're at the end of the
page (accessing the very last guest pde) we can even overflow into the
next host page and oops.
Signed-off-by: Avi Kivity <avi@qumranet.com>
failed VM entry on VMX might still change %fs or %gs, thus make sure
that KVM always reloads the segment selectors. This is crutial on both
x86 and x86_64: x86 has __KERNEL_PDA in %fs on which things like
'current' depends and x86_64 has 0 there and needs MSR_GS_BASE to work.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Intel virtualization extensions do not support virtualizing real mode. So
kvm uses virtualized vm86 mode to run real mode code. Unfortunately, this
virtualized vm86 mode does not support the so called "big real" mode, where
the segment selector and base do not agree with each other according to the
real mode rules (base == selector << 4).
To work around this, kvm checks whether a selector/base pair violates the
virtualized vm86 rules, and if so, forces it into conformance. On a
transition back to protected mode, if we see that the guest did not touch
a forced segment, we restore it back to the original protected mode value.
This pile of hacks breaks down if the gdt has changed in real mode, as it
can cause a segment selector to point to a system descriptor instead of a
normal data segment. In fact, this happens with the Windows bootloader
and the qemu acpi bios, where a protected mode memcpy routine issues an
innocent 'pop %es' and traps on an attempt to load a system descriptor.
"Fix" by checking if the to-be-restored selector points at a system segment,
and if so, coercing it into a normal data segment. The long term solution,
of course, is to abandon vm86 mode and use emulation for big real mode.
Signed-off-by: Avi Kivity <avi@qumranet.com>
PAGE_MASK is an unsigned long, so using it to mask physical addresses on
i386 (which are 64-bit wide) leads to truncation. This can result in
page->private of unrelated memory pages being modified, with disasterous
results.
Fix by not using PAGE_MASK for physical addresses; instead calculate
the correct value directly from PAGE_SIZE. Also fix a similar BUG_ON().
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Avi Kivity <avi@qumranet.com>
KVM shadow page tables are always in pae mode, regardless of the guest
setting. This means that a guest pde (mapping 4MB of memory) is mapped
to two shadow pdes (mapping 2MB each).
When the guest writes to a pte or pde, we intercept the write and emulate it.
We also remove any shadowed mappings corresponding to the write. Since the
mmu did not account for the doubling in the number of pdes, it removed the
wrong entry, resulting in a mismatch between shadow page tables and guest
page tables, followed shortly by guest memory corruption.
This patch fixes the problem by detecting the special case of writing to
a non-pae pde and adjusting the address and number of shadow pdes zapped
accordingly.
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Avi Kivity <avi@qumranet.com>
