This series of patches provides an interface by which userspace can
read and write the hashed page table (HPT) of a Book3S HV guest.
The interface is an ioctl which provides a file descriptor which can
be accessed with the read() and write() system calls. The data read
and written is the guest view of the HPT, in which the second
doubleword of each HPTE (HPT entry) contains a guest physical address,
as distinct from the real HPT that the hardware accesses, where the
second doubleword of each HPTE contains a real address.
Because the HPT is divided into groups (HPTEGs) of 8 entries each,
where each HPTEG usually only contains a few valid entries, or none,
the data format that we use does run-length encoding of the invalid
entries, so in fact the invalid entries take up no space in the
stream.
The interface also provides for doing multiple passes over the HPT,
where the first pass provides information on all HPTEs, and subsequent
passes only return the HPTEs that have changed since the previous pass.
I have implemented a read/write interface rather than an mmap-based
interface because the data is not stored contiguously anywhere in
kernel memory. Of each 16-byte HPTE, the first 8 bytes come from the
real HPT and the second 8 bytes come from the parallel vmalloc'd array
where we store the guest view of the guest physical address,
permissions, accessed/dirty bits etc. Thus a mmap-based interface
would not be practicable (not without doubling the size of the
parallel array, typically requiring an extra 8MB of kernel memory per
guest). This is also why I have not used the memslot interface for
this.
This implements the interface for HV-style KVM but not for PR-style
KVM. Userspace does not need any additional interface with PR-style
KVM because userspace maintains the guest HPT already in that case,
and has an image of the guest view of the HPT in its address space.
This series is against the next branch of the kvm tree. The patches
are basically identical to the previous posting of the series, just
rediffed for the move of kvm.h from include/linux to
include/uapi/linux, and for commit 8ca40a70a7 ("KVM: Take kvm instead
of vcpu to mmu_notifier_retry"), which supersedes patch 1 of the old
series.
The overall diffstat is:
Documentation/virtual/kvm/api.txt | 53 +++++
arch/powerpc/include/asm/kvm_book3s.h | 8 +-
arch/powerpc/include/asm/kvm_book3s_64.h | 24 ++
arch/powerpc/include/asm/kvm_host.h | 1 +
arch/powerpc/include/asm/kvm_ppc.h | 2 +
arch/powerpc/include/uapi/asm/kvm.h | 24 ++
arch/powerpc/kvm/book3s_64_mmu_hv.c | 380 +++++++++++++++++++++++++++++-
arch/powerpc/kvm/book3s_hv.c | 12 -
arch/powerpc/kvm/book3s_hv_rm_mmu.c | 71 ++++--
arch/powerpc/kvm/powerpc.c | 17 ++
include/uapi/linux/kvm.h | 3 +
11 files changed, 551 insertions(+), 44 deletions(-)
Please apply.
Paul.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
