qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v2 00/12] Portable thread-pool/AIO, Win32 emulat


From: Anthony Liguori
Subject: Re: [Qemu-devel] [PATCH v2 00/12] Portable thread-pool/AIO, Win32 emulated AIO
Date: Thu, 25 Oct 2012 13:48:35 -0500
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:11.0) Gecko/20120329 Thunderbird/11.0.1

On 10/25/2012 11:23 AM, Paolo Bonzini wrote:
Il 25/10/2012 18:01, Stefan Hajnoczi ha scritto:
On Tue, Aug 7, 2012 at 1:17 PM, Paolo Bonzini<address@hidden>  wrote:
This patch series is part 2 in my EventNotifier/AIO improvements
for QEMU 1.2.  It extends use of EventNotifier to the main loop
and AIO subsystems.  A new API using EventNotifier is added to aio.c
and a new portable thread pool is introduced (based on code from
posix-aio-compat.c, mostly) that uses this API.  raw-posix.c is
converted to use the new thread pool, and support for asynchronous
I/O is finally added to Win32 as well.

The network drivers (curl, libiscsi, nbd) have to be disabled
under Windows.  They are unlikely to have any users, since they
were broken until 1.0 and (unlike slirp) we never had any report.

I tested this under Wine, with a RHEL virtual machine booting just as
glacially as before.  "info blockstats" does show a slightly higher
overhead, so I would like this to be tested on real Windows hosts.
Even if the result is negative, I would prefer to keep the early
parts (i.e. drop only the last patch) since they are a prerequisite for
improvements to block/raw-posix.c (such as asynchronous discard
support) scheduled for 1.3.  The platform independent APIs introduced
by patches 4 and 5 are also useful for native AIO on Win32.


Paolo Bonzini (12):
   event_notifier: enable it to use pipes
   event_notifier: add Win32 implementation
   main-loop: use event notifiers
   aio: provide platform-independent API
   aio: add Win32 implementation
   linux-aio: use event notifiers
   qemu-thread: add QemuSemaphore
   aio: add generic thread-pool facility
   block: switch posix-aio-compat to threadpool
   raw: merge posix-aio-compat.c into block/raw-posix.c
   raw-posix: rename raw-posix-aio.h, hide unavailable prototypes
   raw-win32: add emulated AIO support

  Makefile.objs                        |  12 +-
  aio.c =>  aio-posix.c                 |   9 +
  aio-win32.c                          | 177 +++++++++
  block/Makefile.objs                  |   6 +-
  block/{raw-posix-aio.h =>  raw-aio.h} |  19 +-
  block/raw-posix.c                    | 297 ++++++++++++++-
  block/raw-win32.c                    | 187 +++++++---
  event_notifier-posix.c               | 118 ++++++
  event_notifier-win32.c               |  59 +++
  event_notifier.c                     |  67 ----
  event_notifier.h                     |  20 +-
  linux-aio.c                          |  51 ++-
  main-loop.c                          | 106 +-----
  oslib-posix.c                        |  31 --
  posix-aio-compat.c                   | 679 -----------------------------------
  qemu-aio.h                           |  19 +-
  qemu-common.h                        |   1 -
  qemu-thread-posix.c                  |  74 ++++
  qemu-thread-posix.h                  |   5 +
  qemu-thread-win32.c                  |  35 ++
  qemu-thread-win32.h                  |   4 +
  qemu-thread.h                        |   7 +
  thread-pool.c                        | 279 ++++++++++++++
  thread-pool.h                        |  34 ++
  trace-events                         |   5 +
  25 file modificati, 1323 inserzioni(+), 978 rimozioni(-)
  rename aio.c =>  aio-posix.c (92%)
  create mode 100644 aio-win32.c
  rename block/{raw-posix-aio.h =>  raw-aio.h} (62%)
  create mode 100644 event_notifier-posix.c
  create mode 100644 event_notifier-win32.c
  delete mode 100644 event_notifier.c
  delete mode 100644 posix-aio-compat.c
  create mode 100644 thread-pool.c
  create mode 100644 thread-pool.h

Ping, is there a v3?

No, but I rebased it on top of AioContext in my branch.  If you're
interested I can resubmit it.

What was the conclusion with 12/12?

I'm pretty happy with the series. If no one is unhappy about 12/12 then I think we should apply it.

Regards,

Anthony Liguori


Paolo





reply via email to

[Prev in Thread] Current Thread [Next in Thread]