- 20 Sep, 2012 6 commits
-
-
Avi Kivity authored
The page table walk is coded as an infinite loop, with a special case on the last pte. Code it as an ordinary loop with a termination condition on the last pte (large page or walk length exhausted), and put the last pte handling code after the loop where it belongs. Reviewed-by:
Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Signed-off-by:
Avi Kivity <avi@redhat.com>
-
Avi Kivity authored
walk_addr_generic() permission checks are a maze of branchy code, which is performed four times per lookup. It depends on the type of access, efer.nxe, cr0.wp, cr4.smep, and in the near future, cr4.smap. Optimize this away by precalculating all variants and storing them in a bitmap. The bitmap is recalculated when rarely-changing variables change (cr0, cr4) and is indexed by the often-changing variables (page fault error code, pte access permissions). The permission check is moved to the end of the loop, otherwise an SMEP fault could be reported as a false positive, when PDE.U=1 but PTE.U=0. Noted by Xiao Guangrong. The result is short, branch-free code. Reviewed-by:
-
Avi Kivity authored
While unspecified, the behaviour of Intel processors is to first perform the page table walk, then, if the walk was successful, to atomically update the accessed and dirty bits of walked paging elements. While we are not required to follow this exactly, doing so will allow us to perform the access permissions check after the walk is complete, rather than after each walk step. (the tricky case is SMEP: a zero in any pte's U bit makes the referenced page a supervisor page, so we can't fault on a one bit during the walk itself). Reviewed-by:
-
Avi Kivity authored
We no longer rely on paging_tmpl.h defines; so we can move the function to mmu.c. Rely on zero extension to 64 bits to get the correct nx behaviour. Reviewed-by:
-
Avi Kivity authored
If nx is disabled, then is gpte[63] is set we will hit a reserved bit set fault before checking permissions; so we can ignore the setting of efer.nxe. Reviewed-by:
-
Avi Kivity authored
gpte_access() computes the access permissions of a guest pte and also write-protects clean gptes. This is wrong when we are servicing a write fault (since we'll be setting the dirty bit momentarily) but correct when instantiating a speculative spte, or when servicing a read fault (since we'll want to trap a following write in order to set the dirty bit). It doesn't seem to hurt in practice, but in order to make the code readable, push the write protection out of gpte_access() and into a new protect_clean_gpte() which is called explicitly when needed. Reviewed-by:
-
- 17 Sep, 2012 5 commits
-
-
Stefan Fritsch authored
- mention that system time needs to be added to wallclock time - positive tsc_shift means left shift, not right - mention additional 32bit right shift Signed-off-by:
Stefan Fritsch <sf@sfritsch.de> Signed-off-by:
Marcelo Tosatti <mtosatti@redhat.com>
-
Michael S. Tsirkin authored
vcpu mutex can be held for unlimited time so taking it with mutex_lock on an ioctl is wrong: one process could be passed a vcpu fd and call this ioctl on the vcpu used by another process, it will then be unkillable until the owner exits. Call mutex_lock_killable instead and return status. Note: mutex_lock_interruptible would be even nicer, but I am not sure all users are prepared to handle EINTR from these ioctls. They might misinterpret it as an error. Cleanup paths expect a vcpu that can't be used by any userspace so this will always succeed - catch bugs by calling BUG_ON. Catch callers that don't check return state by adding __must_check. Signed-off-by:
Michael S. Tsirkin <mst@redhat.com> Signed-off-by:
-
Avi Kivity authored
Use macros for bitness-insensitive register names, instead of rolling our own. Signed-off-by:
-
Avi Kivity authored
Use macros for bitness-insensitive register names, instead of rolling our own. Signed-off-by:
-
Avi Kivity authored
LTO (link-time optimization) doesn't like local labels to be referred to from a different function, since the two functions may be built in separate compilation units. Use an external variable instead. Signed-off-by:
-
- 12 Sep, 2012 1 commit
-
-
Takuya Yoshikawa authored
find_highest_vector() and count_vectors(): - Instead of using magic values, define and use proper macros. find_highest_vector(): - Remove likely() which is there only for historical reasons and not doing correct branch predictions anymore. Using such heuristics to optimize this function is not worth it now. Let CPUs predict things instead. - Stop checking word[0] separately. This was only needed for doing likely() optimization. - Use for loop, not while, to iterate over the register array to make the code clearer. Note that we actually confirmed that the likely() did wrong predictions by inserting debug code. Acked-by:
Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp> Signed-off-by:
-
- 10 Sep, 2012 2 commits
-
-
Xiao Guangrong authored
Checking the return of kvm_mmu_get_page is unnecessary since it is guaranteed by memory cache Signed-off-by:
-
Liu, Jinsong authored
KVM lapic timer and tsc deadline timer based on hrtimer, setting a leftmost node to rb tree and then do hrtimer reprogram. If hrtimer not configured as high resolution, hrtimer_enqueue_reprogram do nothing and then make kvm lapic timer and tsc deadline timer fail. Signed-off-by:
Liu, Jinsong <jinsong.liu@intel.com> Signed-off-by:
-
- 09 Sep, 2012 1 commit
-
-
Jan Kiszka authored
Signed-off-by:
Jan Kiszka <jan.kiszka@siemens.com> Signed-off-by:
-
- 06 Sep, 2012 8 commits
-
-
Michael S. Tsirkin authored
interrupt_bitmap is KVM_NR_INTERRUPTS bits in size, so just use that instead of hard-coded constants and math. Signed-off-by:
-
Gleb Natapov authored
Optimize "rep ins" by allowing emulator to write back more than one datum at a time. Introduce new operand type OP_MEM_STR which tells writeback() that dst contains pointer to an array that should be written back as opposite to just one data element. Signed-off-by:
Gleb Natapov <gleb@redhat.com> Signed-off-by:
-
Gleb Natapov authored
Remove unneeded segment argument. Address structure already has correct segment which was put there during decode. Signed-off-by:
-
Gleb Natapov authored
Signed-off-by:
-
Gleb Natapov authored
Current code assumes that IO exit was due to instruction emulation and handles execution back to emulator directly. This patch adds new userspace IO exit completion callback that can be set by any other code that caused IO exit to userspace. Signed-off-by:
-
Marcelo Tosatti authored
Other arches do not need this. Signed-off-by:
-
Marcelo Tosatti authored
PPC must flush all translations before the new memory slot is visible. Signed-off-by:
-
Marcelo Tosatti authored
Introducing kvm_arch_flush_shadow_memslot, to invalidate the translations of a single memory slot. Signed-off-by:
-
- 05 Sep, 2012 8 commits
-
-
Mathias Krause authored
We never modify direct_access_msrs[], msrpm_ranges[], svm_exit_handlers[] or x86_intercept_map[] at runtime. Mark them r/o. Signed-off-by:
Mathias Krause <minipli@googlemail.com> Cc: Joerg Roedel <joerg.roedel@amd.com> Signed-off-by:
-
Mathias Krause authored
We use vmcs_field_to_offset_table[], kvm_vmx_segment_fields[] and kvm_vmx_exit_handlers[] as lookup tables only -- make them r/o. Signed-off-by:
-
Mathias Krause authored
Signed-off-by:
-
Mathias Krause authored
We never change those, make them r/o. Signed-off-by:
-
Mathias Krause authored
We never change emulate_ops[] at runtime so it should be r/o. Signed-off-by:
-
Mathias Krause authored
The opcode tables never change at runtime, therefor mark them const. Signed-off-by:
-
Mathias Krause authored
As the the compiler ensures that the memory operand is always aligned to a 16 byte memory location, use the aligned variant of MOVDQ for read_sse_reg() and write_sse_reg(). Signed-off-by:
-
Mathias Krause authored
Some fields can be constified and/or made static to reduce code and data size. Numbers for a 32 bit build: text data bss dec hex filename before: 3351 80 0 3431 d67 cpuid.o after: 3391 0 0 3391 d3f cpuid.o Signed-off-by:
-
- 04 Sep, 2012 1 commit
-
-
Gleb Natapov authored
kvm_pic_reset() is not used anywhere. Move reset logic from pic_ioport_write() there. Signed-off-by:
-
- 30 Aug, 2012 1 commit
-
-
Marcelo Tosatti authored
Signed-off-by:
-
- 27 Aug, 2012 7 commits
-
-
Avi Kivity authored
We will enter the guest with G and D cleared; as real hardware ignores D in real mode, and G is taken care of by the limit test, we allow more code to run in vm86 mode. Signed-off-by:
-
Avi Kivity authored
Signed-off-by:
-
Avi Kivity authored
While this is undocumented, real processors do not reload the segment limit and access rights when loading a segment register in real mode. Real programs rely on it so we need to comply with this behaviour. Signed-off-by:
-
Avi Kivity authored
emulate_invalid_guest_state=1 doesn't mean we don't munge the segments in the vmcs; we do. So we need to return the real ones (maintained by vmx_set_segment). Signed-off-by:
-
Avi Kivity authored
We want the segment selector, nor segment number. Signed-off-by:
-
Avi Kivity authored
Segment limits are verified in real mode, not just protected mode. Signed-off-by:
-
Avi Kivity authored
When loading a segment in real mode, only the base and selector must be modified. The limit needs to be left alone, otherwise big real mode users will hit a #GP due to limit checking (currently this is suppressed because we don't check limits in real mode). Signed-off-by:
-
