> On (Mon) Apr 05 2010 [17:33:38], Paul Brook wrote: > > > There might be cases where a few bytes would have been sent out to char > > > devices and some not. Currently the return values from qemu_chr_write() > > > to char devs are only -1, indicating an error, or the complete length > > > of the string passed. > > > > > > Make 'len' a pointer instead, and indicate how much of the string was > > > written. The return value will either be the same as 'len' or a > > > negative number indicating an error condition. > > > > This seems wrong. We should not be getting recoverable errors. > > I was thinking of adding a bool to CharDriverState to indicate if EAGAIN > should be reported to the writer. This can be conveyed at the time of > doing qemu_chr_add_handlers(). > > It would certainly be beneficial for consumers of virtio-serial to be > notified of -EAGAIN so that the guest can be throttled till the chardev > catches up with the data being sent.
EAGAIN should only ever occur if no bytes are written. If a stall condition occurs after some data has been written (and we allow partial completion) then this is indicated by returning a short count. Paul