On 9 July 2018 at 14:17, Paolo Bonzini <pbonz...@redhat.com> wrote: > On 09/07/2018 14:23, Marc-André Lureau wrote: >> I suppose the patch should also change qemu_chr_fe_write() to >> qemu_chr_fe_write_all() to be complete and let the chardev handle >> EAGAIN. > > No, using qemu_chr_fe_write_all() is even more wrong, because hw/char > devices should never block (and hw/char/serial.c handles flow control > properly). > > However, indeed it seems to me that the logic of the patch is backwards: > > - -1/EAGAIN should retry > > - 0 should *not* retry, because it means the other side has hung up
This seems weird, because it doesn't follow the usual pattern for non-blocking functions, where 0 just means "nothing was written" and actual error conditions like the other side having gone away are reported via -1 and some errno. > - everything else should not retry. thanks -- PMM
