On 03.12.2011 16:39, Anssi Hannula wrote: > On 03.12.2011 08:05, Thomas Backlund wrote: >> Anssi Hannula skrev 3.12.2011 01:06: >>> On 03.12.2011 00:18, Thomas Backlund wrote: >>>> Anssi Hannula skrev 2.12.2011 23:17: >>>>> >>>>> Patch attached. I guess this would remain as a Mageia specific patch as >>>>> the wanted behavior is dependant on how the distribution handles driver >>>>> switching etc... >>>> >>>> Speaking of driver switching... >>>> >>>> Is it possible to detect missing radeon firmware in initrd in early >>>> boot, and then switch to loading radeon with modeset=0 ? >>> >>> Seems doable (if a bit hacky), but unfortunately new radeon cards (at >>> least cayman/northern islands) do not work without KMS at all (while >>> slightly older cards like Evergreen mostly work but without any kind of >>> 2d/3d acceleration). >>> >> >> Yeah, I started thinking about those too after my post... >> >>> This also means that X server will fail to start since it doesn't >>> fallback to autodetection (and therefore vesa) when the card is >>> explicitely listed in xorg.conf. >>> >>> Not quite sure what is the best way forward. >>> >> >> Hm, so we would need a split in detection for nonkms capable cards >> and the rest... >> >> or take the easy way out and boot everyone on vesa until radeon firmware >> is available... > > Does loading of radeon modesetting kernel driver without firmware break > the boot completely, or does it just make X.org server fail to start?
Does anyone know? >>> BTW, shouldn't we regenerate initrds when installing radeon-firmware for >>> the first time? >>> >> >> Yeah, I've been thinking the same, but haven't gotten around to doing it >> yet. One possibility is to make dracut record the missing firmware files somewhere, and then in %post or in filetrigger (for /lib/firmware) check if missing firmware got installed, and in that case run bootloader-config --action rebuild-initrds. >> and what about nvidia* and fglrx, should they do the same ? > > Well, currently we always include the KMS drivers in initrd, and when > nvidia* or fglrx is enabled we simply add nokmsboot (mga/mdv specific) > kernel parameter to avoid initrd rebuild. > >> and for a dkms related issue, I've also been thinking we whould >> trigger dkms rebuild on kernel install to save time on next bootup. > > That would seem reasonable, it would just be a matter of > [ -x /usr/sbin/dkms_autoinstaller ] && \ > /usr/sbin/dkms_autoinstaller start %kernelver || : > in kernel %post AFAICS... > -- Anssi Hannula
