Since vmx_get_cpl() always returns 0 when VCPU is in real mode it is no
longer needed. Also reset CPL cache to zero during transaction to
protected mode since transaction may happen while CS.selectors & 3 != 0,
but in reality CPL is 0.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Make fastop opcodes usable in other emulations.
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi.kivity@gmail.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This is a bit of a special case since we don't have the usual
byte/word/long/quad switch; instead we switch on the condition code embedded
in the instruction.
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi.kivity@gmail.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The current reexecute_instruction can not well detect the failed instruction
emulation. It allows guest to retry all the instructions except it accesses
on error pfn
For example, some cases are nested-write-protect - if the page we want to
write is used as PDE but it chains to itself. Under this case, we should
stop the emulation and report the case to userspace
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Currently, reexecute_instruction refused to retry all instructions if
tdp is enabled. If nested npt is used, the emulation may be caused by
shadow page, it can be fixed by dropping the shadow page. And the only
condition that tdp can not retry the instruction is the access fault
on error pfn
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Little cleanup for reexecute_instruction, also use gpa_to_gfn in
retry_instruction
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If the userspace starts dirty logging for a large slot, say 64GB of
memory, kvm_mmu_slot_remove_write_access() needs to hold mmu_lock for
a long time such as tens of milliseconds. This patch controls the lock
hold time by asking the scheduler if we need to reschedule for others.
One penalty for this is that we need to flush TLBs before releasing
mmu_lock. But since holding mmu_lock for a long time does affect not
only the guest, vCPU threads in other words, but also the host as a
whole, we should pay for that.
In practice, the cost will not be so high because we can protect a fair
amount of memory before being rescheduled: on my test environment,
cond_resched_lock() was called only once for protecting 12GB of memory
even without THP. We can also revisit Avi's "unlocked TLB flush" work
later for completely suppressing extra TLB flushes if needed.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Better to place mmu_lock handling and TLB flushing code together since
this is a self-contained function.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
No reason to make callers take mmu_lock since we do not need to protect
kvm_mmu_change_mmu_pages() and kvm_mmu_slot_remove_write_access()
together by mmu_lock in kvm_arch_commit_memory_region(): the former
calls kvm_mmu_commit_zap_page() and flushes TLBs by itself.
Note: we do not need to protect kvm->arch.n_requested_mmu_pages by
mmu_lock as can be seen from the fact that it is read locklessly.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
This makes it possible to release mmu_lock and reschedule conditionally
in a later patch. Although this may increase the time needed to protect
the whole slot when we start dirty logging, the kernel should not allow
the userspace to trigger something that will hold a spinlock for such a
long time as tens of milliseconds: actually there is no limit since it
is roughly proportional to the number of guest pages.
Another point to note is that this patch removes the only user of
slot_bitmap which will cause some problems when we increase the number
of slots further.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
No longer need to care about the mapping level in this function.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Calling kvm_mmu_slot_remove_write_access() for a deleted slot does
nothing but search for non-existent mmu pages which have mappings to
that deleted memory; this is safe but a waste of time.
Since we want to make the function rmap based in a later patch, in a
manner which makes it unsafe to be called for a deleted slot, we makes
the caller see if the slot is non-zero and being dirty logged.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
We have two issues in current code:
- if target gfn is used as its page table, guest will refault then kvm will use
small page size to map it. We need two #PF to fix its shadow page table
- sometimes, say a exception is triggered during vm-exit caused by #PF
(see handle_exception() in vmx.c), we remove all the shadow pages shadowed
by the target gfn before go into page fault path, it will cause infinite
loop:
delete shadow pages shadowed by the gfn -> try to use large page size to map
the gfn -> retry the access ->...
To fix these, we can adjust page size early if the target gfn is used as page
table
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If the write-fault access is from supervisor and CR0.WP is not set on the
vcpu, kvm will fix it by adjusting pte access - it sets the W bit on pte
and clears U bit. This is the chance that kvm can change pte access from
readonly to writable
Unfortunately, the pte access is the access of 'direct' shadow page table,
means direct sp.role.access = pte_access, then we will create a writable
spte entry on the readonly shadow page table. It will cause Dirty bit is
not tracked when two guest ptes point to the same large page. Note, it
does not have other impact except Dirty bit since cr0.wp is encoded into
sp.role
It can be fixed by adjusting pte access before establishing shadow page
table. Also, after that, no mmu specified code exists in the common function
and drop two parameters in set_spte
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Opcodes:
TEST
CMP
ADD
ADC
SUB
SBB
XOR
OR
AND
Acked-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi.kivity@gmail.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
We emulate arithmetic opcodes by executing a "similar" (same operation,
different operands) on the cpu. This ensures accurate emulation, esp. wrt.
eflags. However, the prologue and epilogue around the opcode is fairly long,
consisting of a switch (for the operand size) and code to load and save the
operands. This is repeated for every opcode.
This patch introduces an alternative way to emulate arithmetic opcodes.
Instead of the above, we have four (three on i386) functions consisting
of just the opcode and a ret; one for each operand size. For example:
.align 8
em_notb:
not %al
ret
.align 8
em_notw:
not %ax
ret
.align 8
em_notl:
not %eax
ret
.align 8
em_notq:
not %rax
ret
The prologue and epilogue are shared across all opcodes. Note the functions
use a special calling convention; notably eflags is an input/output parameter
and is not clobbered. Rather than dispatching the four functions through a
jump table, the functions are declared as a constant size (8) so their address
can be calculated.
Acked-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi.kivity@gmail.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
CPL is always 0 when in real mode, and always 3 when virtual 8086 mode.
Using values other than those can cause failures on operations that
check CPL.
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
MMU code tries to avoid if()s HW is not able to predict reliably by using
bitwise operation to streamline code execution, but in case of a dirty bit
folding this gives us nothing since write_fault is checked right before
the folding code. Lets just piggyback onto the if() to make code more clear.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
With emulate_invalid_guest_state=0 if a vcpu is in real mode VMX can
enter the vcpu with smaller segment limit than guest configured. If the
guest tries to access pass this limit it will get #GP at which point
instruction will be emulated with correct segment limit applied. If
during the emulation IO is detected it is not handled correctly. Vcpu
thread should exit to userspace to serve the IO, but it returns to the
guest instead. Since emulation is not completed till userspace completes
the IO the faulty instruction is re-executed ad infinitum.
The patch fixes that by exiting to userspace if IO happens during
instruction emulation.
Reported-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Segment registers will be fixed according to current emulation policy
during switching to real mode for the first time.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Currently when emulation of invalid guest state is enable
(emulate_invalid_guest_state=1) segment registers are still fixed for
entry to vm86 mode some times. Segment register fixing is avoided in
enter_rmode(), but vmx_set_segment() still does it unconditionally.
The patch fixes it.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Currently it allows entering vm86 mode if segment limit is greater than
0xffff and db bit is set. Both of those can cause incorrect execution of
instruction by cpu since in vm86 mode limit will be set to 0xffff and db
will be forced to 0.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
According to Intel SDM Vol3 Section 5.5 "Privilege Levels" and 5.6
"Privilege Level Checking When Accessing Data Segments" RPL checking is
done during loading of a segment selector, not during data access. We
already do checking during segment selector loading, so drop the check
during data access. Checking RPL during data access triggers #GP if
after transition from real mode to protected mode RPL bits in a segment
selector are set.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Segment descriptor's base is fixed by call to fix_rmode_seg(). Not need
to do it twice.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
The code for SS and CS does the same thing fix_rmode_seg() is doing.
Use it instead of hand crafted code.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
VMX without unrestricted mode cannot virtualize real mode, so if
emulate_invalid_guest_state=0 kvm uses vm86 mode to approximate
it. Sometimes, when guest moves from protected mode to real mode, it
leaves segment descriptors in a state not suitable for use by vm86 mode
virtualization, so we keep shadow copy of segment descriptors for internal
use and load fake register to VMCS for guest entry to succeed. Till
now we kept shadow for all segments except SS and CS (for SS and CS we
returned parameters directly from VMCS), but since commit a5625189f6
emulator enforces segment limits in real mode. This causes #GP during move
from protected mode to real mode when emulator fetches first instruction
after moving to real mode since it uses incorrect CS base and limit to
linearize the %rip. Fix by keeping shadow for SS and CS too.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
rmode_segment_valid() checks if segment descriptor can be used to enter
vm86 mode. VMX spec mandates that in vm86 mode CS register will be of
type data, not code. Lets allow guest entry with vm86 mode if the only
problem with CS register is incorrect type. Otherwise entire real mode
will be emulated.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Set segment fields explicitly instead of using binary operations.
No behaviour changes.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
The kvm i8254 emulation for counter 0 (but not for counters 1 and 2)
has at least two bugs in mode 0:
1. The OUT bit, computed by pit_get_out(), is never set high.
2. The counter value, computed by pit_get_count(), wraps back around to
the initial counter value, rather than wrapping back to 0xFFFF
(which is the behavior described in the comment in __kpit_elapsed,
the behavior implemented by qemu, and the behavior observed on AMD
hardware).
The bug stems from __kpit_elapsed computing the elapsed time mod the
initial counter value (stored as nanoseconds in ps->period). This is both
unnecessary (none of the callers of kpit_elapsed expect the value to be
at most the initial counter value) and incorrect (it causes pit_get_count
to appear to wrap around to the initial counter value rather than 0xFFFF).
Removing this mod from __kpit_elapsed fixes both of the above bugs.
Signed-off-by: Nickolai Zeldovich <nickolai@csail.mit.edu>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
