On Oct 15, 2013, at 2:54 PM, Dave Chinner <david@xxxxxxxxxxxxx> wrote: > On Tue, Oct 15, 2013 at 01:33:19PM -0700, Christoph Hellwig wrote: >> Hi Paulo, >> >> just wondering what the state of the xfs support for syslinux is? I >> talked to Peter at Linuxcon and he thought it's merged, but looking at >> the kernel.org tree I can't find the support. >> >> Also when looking over your branches I noticed that you're using a free >> sector in the first filesystem block to store the bootloader. If we >> want to go down that route we need to make sure to reserve this sector, >> otherwise it might get taken up by newly added metadata. > > It's also worth pointing out that there's no guarantee that there's > a free sector in the first filesystem block. It's only by luck that > there's free sectors on the default config (512 byte sector, 4 > sector sized AG headers, 4k filesystem block). If we have <= 2k filesystem > block there are no free "pad" sectors that can be used, 4k sectors > mean no free sectors either, etc. > > Much better would be to create a sector sized file and use fiemap to > get the disk address of the block and feed that into the > bootloader. That works for all filesystems without needing to know > anything about the underlying filesystem structures…… I'm curious how that would work. The minimum bit of code for GRUB or extlinux is a lot bigger than 4KB. A basic self-generated GRUB configuration file is 5.6KB; for extlinux a basic one I have is 518 bytes. The minimum code needed to find the configuration file is ~26KB for GRUB's core.img, and ~34KB for extlinux's ldlinux.sys. So are both of you referring to the < 440 bytes of bootstrap code that goes in either the MBR or VBR, whose job is to find core.img or ldlinux.sys and blindly load them? Chris Murphy _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs