On 03/09/2011 04:33 AM, Jes Sorensen wrote:
On 03/07/11 21:10, Michael Roth wrote:
To be able to use qemu_mod_timer() and friends to register timeout
events for virtagent's qemu-va tool, we need to do the following:
Move several blocks of code out of cpus.c that handle initialization
of qemu's io_thread_fd and working with it via
qemu_notify_event()/qemu_event_read()/etc, and make them accessible
as backend functions to both the emulator code and qemu-tool.c via
wrapper functions within cpus.c and qemu-tool.c, respectively. These
have been added to qemu-ioh.c, where similar treatment was given to
qemu_set_fd_handler() and friends.
Some of these wrapper functions lack declarations when being
built into tools, so we add those via qemu-tool.h, which can be included
by a tool to access them. With these changes we can drive timers in a
tool linking it against qemu-timer.o and then implementing something
similar to the main i/o loop in vl.c:
[snip]
diff --git a/qemu-ioh.c b/qemu-ioh.c
index cc71470..5c3f94c 100644
--- a/qemu-ioh.c
+++ b/qemu-ioh.c
@@ -113,3 +117,94 @@ void qemu_process_fd_handlers2(void *ioh_record_list,
const fd_set *rfds,
}
}
}
+
+#ifndef _WIN32
+void iothread_event_increment(int *io_thread_fd)
Please move these functions into posix/w32 specific files so we don't
get anymore ugly #ifdefs. It would be good if we could use a wrapper
struct as well to hide the different data types so we don't need #ifdefs
in the calling code as well.
Yup, meant to add this to the TODO. I may end up sending these general
tools changes out in a separate patchset since they seem to be in
conflict with quite of few patches floating around the list. Either way
I'll make sure to get these cleaned up and tested a bit a more.
Cheers,
Jes
