Permalink
Commits on Aug 30, 2013
  1. merge pre-ept branch

    estibi committed Aug 29, 2013
Commits on Mar 21, 2013
Commits on Nov 27, 2012
Commits on Oct 23, 2012
Commits on Sep 24, 2012
Commits on Aug 27, 2012
Commits on Aug 9, 2012
Commits on Jul 16, 2012
Commits on Jun 28, 2012
Commits on Jun 20, 2012
  1. OS-1298 kvm builds mdb module against the build system's headers

    Keith M Wesolowski
    Keith M Wesolowski committed Jun 19, 2012
    OS-1299 kvm is still using Studio
    OS-1314 kvm.so is wrongly built as if it were a kmod
Commits on Mar 8, 2012
Commits on Mar 4, 2012
Commits on Mar 3, 2012
  1. missed a kmem_free in cc6efc3

    jclulow committed Mar 3, 2012
Commits on Mar 1, 2012
Commits on Jan 27, 2012
  1. handle required preemption in generic x86 layer

    jclulow committed Sep 25, 2011
    We have been fudging attention from the dispatcher in the svm guest cpu run
    code by setting kprunrun after a #VMEXIT.  It would be better to preempt only
    if the userland runrun flag (or kprunrun) is set already, avoiding unecessary
    switches.  It would better again to do this outside of the instruction
    set-specific code so we'll do it in the generic x86 guest cpu run loop,
    exiting back to userland if we should get off the CPU.
    
    Note that the joyent vmx code already exits like this in an exit handler,
    which I have marked with a comment.  We should test to see if we can remove
    this from vmx, deferring to kvm_x86 to induce the exit.
  2. Re-enable optimization

    richlowe authored and jclulow committed Sep 17, 2011
  3. svm: Allocate IOPM physically contiguously

    richlowe authored and jclulow committed Sep 17, 2011
    The IOPM resides in 3 pages which _must_ be physically contiguous, as
    they're referenced by the CPU.  Allocate them with contig_alloc(), which
    we really shouldn't be using.
    
    This appears to prevent both the rtls reset failure and the ATA
    timeouts, as both were using I/O ports which would fall into the second
    page of the IOPM which, when not physically contiguous, is essentially
    random, meaning that a random selection of IO port accesses were never
    intercepted.
  4. svm: Don't whine about MSR_IA32_DEBUGCTL

    richlowe authored and jclulow committed Sep 17, 2011
  5. x86: Hackily allow MCG writes of non-0/~0

    richlowe authored and jclulow committed Sep 17, 2011
    The SunOS Opteron CPU module only writes 1 bits to the MCG for banks
    physically present in the system, failing this test and injecting a
    fault (which then crashes with an unaligned stack, for possibly separate
    reasons)