http://sn.1w6.org/file/mk-20140627T095123-sfu3uza.html

@mcscx I remember low-level tools under DOS (MS or DR) could reveal these retries; for a floppy the standard was up to 5 attempts before the OS would get a signal: these tools would talk directly to the controller, I think. Then there were (still are) repair tools like SpinRite: often the cause of a bad sector is simply too weak magnetization, and this may be repaired by (repeated) rewrites - read a sector and then repeatedly write it back to 'refresh' the magnetization - of course, also at HW sector level. I had a copy of SpinRite and regularly used it on my dodgy HD. I wonder if I could reproduce something like that with Linux tools... Anyway, I doubt dmesg is low-level enough to reveal the re-reads.