Atsushi Nemoto wrote: > On Mon, 16 Oct 2006 10:48:37 +0200, Franck Bui-Huu <vagabon.xyz@xxxxxxxxx> wrote: >> thanks but it doesn't explain anything either...Anyways what about this >> patch on top of the previous one ? > >> + initrd_header = __va(PAGE_ALIGN(__pa_symbol(&_end) + sizeof(u32) * 2 + 1)); > > This breaks the addinitrd. You mean this perhaps? > > initrd_header = __va(PAGE_ALIGN(__pa_symbol(&_end) + sizeof(u32) * 2)) - sizeof(u32) * 2; > you're right, but I really don't see how this work and IMHO this code is broken. In my mind, I was thinking that initrd_head had to be PAGE_SIZE aligned and that was the reason of that weird code... > > BTW, I'm a bit uncomfortable with current automatic initrd detection. > Now we have rd_start= option. If I enabled BLK_DEV_INITRD and did > pass nfsroot= instead of rd_start= option, I want kernel do not search > initrd_header at all. Note that in this case current kernel might > misdetect initrd_header from garbage beyond "_end". > Well that might happen if you want a nfs rootfs but want to execute an initrd before mounting the rootfs and this initrd has been included in the kernel image with the 'addinitrd' stuff. > I think something like CONFIG_INITRD_AUTODETECT to control this > behaviour is useful. What do you think? > It's safer although it can be enough to check against a magic number well chosen. Maybe we can introduce a new option to the command line, 'initrd_noprobe' for example. But in any case make the default to auto detect initrd to avoid breaking some old configs. Franck
