On Wed, Aug 16, 2023 at 11:57:17AM -0300, Fabiano Rosas wrote: > I'm talking about the actual yank action, not the unregister. > > migration_yank_iochannel() calls qio_channel_shutdown() in the same way > as qemu_file_shutdown(), but unlike the latter, it doesn't set > f->last_error = -EIO. Which means that in theory, we could yank and > still try to use the QEMUFile. > > In other words, what commit a555b8092a ("qemu-file: Don't do IO after > shutdown") did does not apply to yank because yank didn't exit at the > time.
Ah ok.. Perhaps we should register yank over the qemufiles not ioc for migrations? -- Peter Xu