Re: [Xen-devel] [PATCH v3.2] libxl: Design of an async API to issue QMP commands to QEMU [and 1 more messages]

[Ian Jackson <ian.jackson@xxxxxxxxxx>]

Anthony PERARD writes ("[PATCH v3.2] libxl: Design of an async API to issue QMP 
commands to QEMU"):
> All the functions will be implemented in later patches.
> 
> This patch includes the API that libxl can use to send QMP commands to
> QEMU.
...
> + * This facility allows a command to be sent to QEMU, and the response to be
> + * handed to a callback function.  Each libxl__ev_qmp handles zero or one
> + * outstanding command; if multiple commands are to be sent concurrently,
> + * multiple libxl__ev_qmp's must be used.
> + *
> + * Possible states:
> + *  Undefined
> + *    Might contain anything.
> + *  Idle
> + *    Struct contents are defined enough to pass to any libxl__ev_qmp_*
> + *    functions.
> + *    The struct does not contain references to any allocated private 
> resources
> + *    so can be thrown away.
> + *  Active
> + *    Currently waiting for the callback to be called.
> + *    _dispose must be called to reclaim resources (or wait for the callback 
> to
> + *    be called).
> + *
> + * libxl__ev_qmp_init: Undefined/Idle -> Idle
> + *
> + * libxl__ev_qmp_send: Idle -> Active (on error: Idle)
> + *    Sends a command to QEMU.
> + *
> + * libxl__ev_qmp_dispose: Active/Idle -> Idle
> + *
> + * callback:
> + *    When called, ev is Idle, so can be reused or thrown away.
> + *    When an error occured, it is called with response == NULL and the error
> + *    code in rc.

You have removed this text:

  + * When called from within a callback, the same QMP connection will be
  + * reused to execute the new command. This is important in the case
  + * where the first command is "add-fd" and the second command use the
  + * fdset created by QEMU.

That removes the need for a fourth state, as I discussed earlier.  But
does this not introduce the problem that this text was addressing: Ie
there is no way for the user of libxl_ev_qmp to ensure that the
commands to add-fd, and use the fdset, occur on the same qmp
connection.

What has changed since you wrote that ?  Or have I misunderstood
something ?

NB that I haven't yet read in detail the part about describing the
implementation, ie the qmp_state but I think this problem exists no
matter what that implementation looks like...

Anthony PERARD writes ("Re: [PATCH v3.2] libxl: Design of an async API to issue 
QMP commands to QEMU"):
> On Mon, Jul 16, 2018 at 04:28:00PM +0100, Anthony PERARD wrote:
> > + * Implementation of the QMP client.
> > + *
> 
> Here, I wanted to add more description, but forgot to commit before to
> sent the patch, it should read:
> 
> + * This struct qmp_state is used by the libxl__ev_qmp_* functions, but it is
> + * not visible to users of libxl__ev_qmp_*. It is allocated as needed and
> + * stored in CTX in order to allow reuse of a same QMP connection between
> + * different users.

This makes sense, thanks.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel