[PATCH RFC 0/7] move softmmu options processing from os-posix.c to vl.c

Michael Tokarev posted 7 patches 9 months, 1 week ago
Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/cover.1691843673.git.mjt@tls.msk.ru
Maintainers: Paolo Bonzini <pbonzini@redhat.com>, Stefan Weil <sw@weilnetz.de>
There is a newer version of this series
include/qemu/qemu-options.h        |  41 --------
include/sysemu/os-posix.h          |   8 +-
include/sysemu/os-win32.h          |   1 -
os-posix.c                         | 152 +++++++++--------------------
{util => softmmu}/async-teardown.c |   0
softmmu/meson.build                |   1 +
softmmu/vl.c                       |  87 ++++++++++++++++-
util/meson.build                   |   1 -
8 files changed, 134 insertions(+), 157 deletions(-)
delete mode 100644 include/qemu/qemu-options.h
rename {util => softmmu}/async-teardown.c (100%)
[PATCH RFC 0/7] move softmmu options processing from os-posix.c to vl.c
Posted by Michael Tokarev 9 months, 1 week ago
qemu_init() calls os_parse_cmd_args(), which is obviously a very vl.c-specicic
piece of code.  It looks like when moving vl.c to softmmu/, os-posix.c should've
been moved too (together with os-win32.c).  But there are other functions in
os-posix.c which are used by other parts of the code, eg qemu-nbd or
qemu-storage-daemon both uses parts from there.  The result is rather ugly.
One of the side-effects is that even if system build is disabled, we still
have to compile util/async-teardown.c, since it is used in os-posix.c, - and
this one can not be built on ia64 due to lack of clone syscall.  But qemu-nbd
does not need async-teardown, which is also very softmmu-specific.

This patchset tries to address these defects.  It moves the "bottom half" of
vl.c options processing back into vl.c, effectively undoing commit 59a5264b99434
from 2010.  We do not have many os-specific options, and this move makes
qemu_init() the only function which processes options, all in the single
place, which is, in my opinion, easier to manage.

Before this move, lower-level constructs (chroot, runas) are being exported.

In the end, we have almost no softmmu-related things in os-posix.c, and the
same functionality can be actually used in qemu-storage-daemon or qemu-nbd.
For qemu-nbd, it is now possible to switch to using -runas/-chroot/etc the
same way as qemu-storage-daemon does, instead of having one more abstraction
named qemu_daemon().

Michael Tokarev (7):
  include/sysemu/os-posix.h: move *daemonize* declaration together
  os-posix: create and export os_set_runas()
  os-posix.c: create and export os_set_chroot()
  os-posix.c, softmmu/vl.c: move os_parse_cmd_args() into qemu_init()
  softmmu/vl.c: inline include/qemu/qemu-options.h into vl.c
  os-posix.c: move code around
  util/async-teardown.c: move to softmmu/, only build it when system
    build is requested

 include/qemu/qemu-options.h        |  41 --------
 include/sysemu/os-posix.h          |   8 +-
 include/sysemu/os-win32.h          |   1 -
 os-posix.c                         | 152 +++++++++--------------------
 {util => softmmu}/async-teardown.c |   0
 softmmu/meson.build                |   1 +
 softmmu/vl.c                       |  87 ++++++++++++++++-
 util/meson.build                   |   1 -
 8 files changed, 134 insertions(+), 157 deletions(-)
 delete mode 100644 include/qemu/qemu-options.h
 rename {util => softmmu}/async-teardown.c (100%)

-- 
2.39.2
Re: [PATCH RFC 0/7] move softmmu options processing from os-posix.c to vl.c
Posted by Eric Blake 8 months, 2 weeks ago
On Sat, Aug 12, 2023 at 03:47:54PM +0300, Michael Tokarev wrote:
> qemu_init() calls os_parse_cmd_args(), which is obviously a very vl.c-specicic
> piece of code.  It looks like when moving vl.c to softmmu/, os-posix.c should've
> been moved too (together with os-win32.c).  But there are other functions in
> os-posix.c which are used by other parts of the code, eg qemu-nbd or
> qemu-storage-daemon both uses parts from there.  The result is rather ugly.
> One of the side-effects is that even if system build is disabled, we still
> have to compile util/async-teardown.c, since it is used in os-posix.c, - and
> this one can not be built on ia64 due to lack of clone syscall.  But qemu-nbd
> does not need async-teardown, which is also very softmmu-specific.
> 
> This patchset tries to address these defects.  It moves the "bottom half" of
> vl.c options processing back into vl.c, effectively undoing commit 59a5264b99434
> from 2010.  We do not have many os-specific options, and this move makes
> qemu_init() the only function which processes options, all in the single
> place, which is, in my opinion, easier to manage.
> 
> Before this move, lower-level constructs (chroot, runas) are being exported.
> 
> In the end, we have almost no softmmu-related things in os-posix.c, and the
> same functionality can be actually used in qemu-storage-daemon or qemu-nbd.
> For qemu-nbd, it is now possible to switch to using -runas/-chroot/etc the
> same way as qemu-storage-daemon does, instead of having one more abstraction
> named qemu_daemon().

Given that qemu-nbd is affected, I'm happy to queue this through my
NBD tree if no one else picks it up first.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.
Virtualization:  qemu.org | libguestfs.org