Hello,
this is the latest version of the changes to KVM MMU code to support 1gb
pages. Pages with a size of 1gb are a feature of all Fam10h AMD CPUs
and this patch set makes them available for guests and allows to map the
nested page table with 1gb pages.
Changes to the previous version of this patch set:
* fixed a bug in the pte reserved bit checking code which
disallowed 1gb ptes (thanks Marcelo)
* included a working version of 1gb page support for shadow
paging (again, thanks to Marcelo who found the bug in the
original patch)
* rebased patches to the latest avi/master branch
I also gathered some new performance data for this code. The following
table contains the data. The configrations are all named as
<1gb pages enabled in guest>-<host backing size>. The numbers below are
the average of the elapsed time of 3 kernel compiles in a guest with 3GB
memory and 4 vpcus on an Istanbul based machine. The make jobs ran with
-j8. All measurements were done with nested paging enabled.
Configuration Elapsed Time Percent
------------------------------------------------------------
no1gbpages-4kb 139.81 100%
no1gbpages-1gb 129.89 92.9%
1gbpages-4kb 137.09 98.0%
1gbpages-1gb 129.42 92.6%
the biggest benefit is gained using 1gb pages for the nested page table.
The performance improvement compared to 4kb backing size with nested
paging is between 7% and 7.5%. Using 1gb pages in a guest with a 4kb
backing size on the host gives around 2% performance improvement.
Please review and/or consider to apply these patches.
Thanks,
Joerg
Diffstat:
arch/x86/include/asm/kvm_host.h | 5 +-
arch/x86/kvm/mmu.c | 242 ++++++++++++++++++++++++---------------
arch/x86/kvm/paging_tmpl.h | 97 ++++++++--------
arch/x86/kvm/svm.c | 6 +
arch/x86/kvm/vmx.c | 6 +
arch/x86/kvm/x86.c | 3 +-
6 files changed, 214 insertions(+), 145 deletions(-)
Shortlog:
Joerg Roedel (7):
kvm/mmu: make rmap code aware of mapping levels
kvm/mmu: rename is_largepage_backed to mapping_level
kvm/mmu: make direct mapping paths aware of mapping levels
kvm/mmu: make page walker aware of mapping levels
kvm/mmu: shadow support for 1gb pages
kvm/mmu: enable gbpages by increasing nr of pagesizes
kvm x86: report 1GB page support to userspace
--
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
