On 18 April 2014 18:08, Kirill Batuzov <batuz...@ispras.ru> wrote: > And something like this happens every time qemu_allocate_irqs is called.
Which generally only happens in the board init stage, ie once. So why care? Leaks are interesting if they happen in code that could be called multiple times (eg for hotplugged devices). A handful of bytes which we only allocate once is irrelevant: it gets freed on QEMU exit anyway. As I say, we will at some point want to overhaul the qemu_irq APIs to more QOM like behaviour, which gets us a bunch of nice things like properly named inputs and outputs and maybe ability to wire things up via config file or command line; and as a side effect we should end up with the corresponding allocations being tracked. But until we get to that, there really isn't a practical problem with the current approach for board level IRQ wiring. If we're leaking IRQ related memory in hotpluggable device init/deinit that might be worth fixing. thanks -- PMM
