On 08/19/2010 11:05 AM, Gerd Hoffmann wrote:
While not really unsafe, the qemu_malloc functions are not guaranteed to
be re-entrant by the interfaces today. It's also terribly subtle to have
to rely on implicit re-entrance safety.
The underlying malloc() is re-entrant, isn't it?
pflib (which is called too) is re-entrant too.
Otherwise only private data is accessed (under lock when needed).
Yes, I looked too and agree that it's safe now. But we're sloppy as
hell in qemu about depending on global state and I can imagine someone
adding something to these functions that would create an issue.
I think documentation would be sufficient, at least for now.
Having spice as an independent interface to the current display_type
switching seems awkward to me.
Having remote desktop protocols as DT_something seems awkward to me.
It makes sense for the local display (being none, curses, sdl, fbdev,
whatever). For remote display protocols I see no reason why we
shouldn't have multiple of them enabled at the same time, so the user
can connect with whatever he wants. And that even in parallel to a
local display if needed.
The state the patch introduces is a bit inconsistent though. But I'd
rather drop DT_VNC instead of adding DT_SPICE.
Yes, I would think that would be reasonable.
Regards,
Anthony Liguori
cheers,
Gerd
