Re: [Qemu-devel] [PATCH v4 4/4] fsdev: QMP interface for throttling

[Greg Kurz]

On Thu, 18 May 2017 15:30:06 +0200
Pradeep Jagadeesh <address@hidden> wrote:

> On 5/17/2017 7:09 PM, Eric Blake wrote:
> > On 05/17/2017 11:29 AM, Greg Kurz wrote:
> >  
> >>>
> >>> First point: is fsdev a Linux-only feature, or can it be compiled on
> >>> BSD?  If it is Linux-only, then compiling a stub for Windows will still
> >>> leave BSD broken, and your #ifdef is wrong.  Fixing compilation on mingw
> >>> is nice, but not the only platform to worry about.
> >>>  
> >>
> >> fsdev compilation currently depends on CONFIG_VIRTFS which is a Linux-only
> >> feature for the moment. There was a tentative to support it on Windows 
> >> hosts
> >> two years ago but it stayed at the RFC stage.
> >>
> >> But even on Linux hosts, the current fsdev implementation also depends on
> >> the target supporting PCI and VIRTIO. We have a fsdev/qemu-fsdev-dummy.c
> >> file to put stubs so that we don't pull all the code for such targets.
> >>
> >> Maybe this could be reused for the above stubs as well ?  
> >
> > That helps.  The stub should live in qemu-fsdev-dummy.c (where it
> > shouldn't need any #ifdef, because that file is only compiled when the
> > condition is false), and...
> >  
> >>  
> >>> Second point: if fsdev is indeed a platform-specific feature, then we
> >>> don't want to advertise the QMP commands that are useless when running
> >>> on a platform that doesn't support it. Anywhere you have to add a stub
> >>> for compilation means you ALSO need to patch monitor.c to unregister the
> >>> command from being advertised as a valid QMP command.  (If you used
> >>> #ifdef __LINUX__ to guard the working version, and #ifndef __LINUX__ to
> >>> declare the stub, then the monitor.c needs an #ifndef section within
> >>> qmp_unregister_commands_hack() to tell QMP to not advertise the stubs.)  
> >
> > monitor.c should wrap the unregister under #ifndef CONFIG_VIRTFS (rather
> > than a particular platform name).  
> I did unregister the functions as mentioned above in monitor.c. But it 
> does not address the issue when I run "make address@hidden".

What issue is this ?

I cannot even run this with your unmodified series (please note that I
always build out-of-tree).

> I think the VIRTFS is still enabled.
> Only having dummy functions even in qmp.c addresses the issue.
> 
> Regards,
> Pradeep
> >  
> 
>

pgpycT17PnKop.pgp
Description: OpenPGP digital signature