John Paul Adrian Glaubitz, on dim. 01 oct. 2017 12:49:25 +0200, wrote: > On 10/01/2017 12:33 PM, Samuel Thibault wrote: > >> Yes, I'm aware of that and that's what people usually use to enable the > >> serial > >> console post installation. However, it's most likely not what we would > >> want to > >> use as adding it to the kernel command line assumes that you know what type > >> of bootloader is being used and you would have to add case-matching for the > >> various bootloader types. > > > > Well, debian-installer already needs to know about that, to be able to > > pass to the installed system the parameters which were given to boot the > > installer itself. > > But is that information available in finish-installation.d/90console?
The point is: it doesn't need to be. We need 90console for non-systemd systems because sysvinit etc. are not able to automatically get the console configuration from the kernel commande line. Systemd, however, is, and thus doesn't need anything in 90console. > >> Please be aware that I am not talking about just fixing the serial console > >> issue > >> on SPARC machines but on all architectures supported by Debian. Hence, I > >> want > >> to use a generic and robust solution, not a quick hack. > > > > Well, your solution is somehow a "quick hack": enabling serial console > > automatically already works "for all archs", there is just something > > which doesn't happen to work in your case, and AIUI it's writing on the > > installed system the kernel parameters which were passed to boot d-i. > > Are you sure that's the case? Yes. > The code to enable the serial console for systemd systems is > definitely missing. Yes, on purpose: it doesn't need to be. > It's there for upstart and sysvinit but not for systemd. By default, > systemd does not start a serial getty unless you pass console=ttyS0 on > the kernel command line or enable the service explicitly. Yes, that's on purpose: it has to be there on the command line for kernel messages to get printed on ttyS0 anyway. > Are you getting a serial console enabled by default on a fresh systemd > installation? What do you mean by "serial console enabled by default"? Perhaps that's where there is misunderstanding. > > Or is there really no way on your system for d-i to be able to write > > down the parameters to be passed to the kernel for the installed system? > > Well, I have just adopted this particular script for systemd and effectively > used the same mechanism as for sysvinit and upstart. Would you have preferred > a completely different mechanism? I don't "prefer" something, I'm just saying how things are currently supposed to work: somehow the kernel for d-i has console=ttyS0 on its command line so that d-i gets its output on ttyS0, and d-i normally copies what is passed to its kernel command line into the installed system, so that the installed system behaves just like d-i behaved, which is normally what the user expects. > There is nothing special about sparc64 in this regard. Perhaps that "enabled by default" thing. Samuel
