Hi -

Our VMM was doing a lot of unnecessary mmaps and munmaps.  The latter
trigger TLB shootdowns.  This patchset cuts out most of the mmap/munmaps
associated with task thread creation (every guest core timer tick) and
eventfd_write (virtio-mmio kicks).  

As a side note, we don't need to use eventfds in virtio - we should use
uthread sync directly and avoid the kernel.

Barret


The following changes since commit 9db45c3e6abf403a0bfe10ead3950d086e15c827:

  Fix matrix bug in gethostbyname (XCC) (2017-07-30 12:45:09 -0700)

are available in the git repository at:

  [email protected]:brho/akaros.git vmm

for you to fetch changes up to 35cc026dc354164086324e24a5502aa27c6feacf:

  eventfd: Use write_hex_to_fd() (XCC) (2017-07-31 16:23:25 -0400)

----------------------------------------------------------------
View this online at:
https://github.com/brho/akaros/compare/9db45c3e6abf...35cc026dc354

----------------------------------------------------------------
Barret Rhoden (5):
      x86: Don't enter the monitor for invalid opcode
      parlib: slab: Use the modern ctor/dtor interface
      parlib: slab: mmap with PROT_EXEC
      vmm: Use a task_thread cache
      eventfd: Use write_hex_to_fd() (XCC)

 kern/arch/x86/trap.c                               |  5 ---
 .../glibc-2.19-akaros/sysdeps/akaros/eventfd.c     |  9 +---
 user/iplib/epoll.c                                 |  7 ++--
 user/parlib/dtls.c                                 |  4 +-
 user/parlib/include/parlib/slab.h                  | 10 +++--
 user/parlib/slab.c                                 | 48 ++++++++++++++--------
 user/vmm/sched.c                                   | 48 +++++++++++++++++-----
 7 files changed, 81 insertions(+), 50 deletions(-)

-- 
You received this message because you are subscribed to the Google Groups 
"Akaros" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to