Yes, dmesg output or any other capture of the full set of guest kernel messages. In particular I'd like to understand whether post-resume the parameters of the block interface changed in any way - for now I could only halfway explain the behavior if some of them shrunk, i.e. the total I/O size that fits on the ring would be less than before suspend. And yes, I realize that this may not be readily reproducible (or else I might be able to debug this locally). If you find a way to halfway reliably reproduce this, would you be able to use debugging or testing patches (i.e. rebuild the kernel yourself)?