Darren J Moffat wrote: > Riny Qian wrote: > >> BTW, kernel messages will always go to the current active virtual >> console. > > > Thats what I said I didn't want. There should in my opinion only be one > /dev/console (modulo what you do with consadm(1m). This project > shouldn't in my opinion be about virtualising /dev/console but instead > about proving additional tty's that are available from the machines > primary keyboard mouse and screen; they are accessed via a magic key > sequence. I believe that is what the expectation is on other platforms. > >
Kernel messages output (e.g. via cmn_err) is different with user level messages output via /dev/console. At user level, firstly you echo "a large file" > /dev/console, the content of that file will go to current virtual console (e.g. /dev/vt1). Then your switch to another virtual console (e.g. /dev/vt2), then you echo "another large file" > /dev/console, the content of it will still go to current virtual console, that is, /dev/vt2. In kernel, each time cmn_err is called, the output will go to current virtual console. There is no way for console subsystem to combine or distinguish different cmn_errs. Or perhaps we can still take the /dev/console policy of older Solaris (e.g. 2.5) with virtual console functionality, under that, /dev/console was the primary console, and VT just provided additional virtual console terminals. Everything that redirects to /dev/console will not go to the current virtual console. Even on such older Solaris, kernel messages still always went through the current virtual console exactly as what we do now. thanks, Riny
