Bdale, I found out today that the d-i hppa netboot images for 2.6 have been broken for a while. I thought they were working because my daily test setup was falling back to using a boot image that I built by hand, probably months ago, because I had a wrong path set. When I fixed it to use the current images, I got this: attempt to access beyond end of device ram0: rw=0, want=18146, limit=16384 attempt to access beyond end of device ram0: rw=0, want=18904, limit=16384 I guess the ramdisk is the wrong size. I'm suprised this didn't come up in your recent testing though -- could it somehow only affect my a500? I tried passing a ramdisk_size=20000 to the kernel and that works around it. However I then don't have enough command-line space to do preseeding. Now it seems to me that the boot.img is 13 mb only because 9.5 mb are used by the kernels. The initrd is only 3.4 mb, and though it uncompresses to a 12 mb image, that doesn't explain how we're filling up a 16 mb ramdisk. Is palo doing something stupid, such as loading the kernels into memory along with the initrd as the ramdisk? Anyway, I notice there's a fair bit of cruft on the hppa initrds and perhaps we should apply this patch: --- pkg-lists/netboot/hppa.cfg (revision 25106) +++ pkg-lists/netboot/hppa.cfg (working copy) @@ -1,10 +1,5 @@ -cdrom-core-modules-${kernel:Version} -ide-modules-${kernel:Version} input-modules-${kernel:Version} nic-modules-${kernel:Version} * -scsi-modules-${kernel:Version} -usb-modules-${kernel:Version} -usb-storage-modules-${kernel:Version} cdebconf-text-udeb console-keymaps-at That stuff used to be included from the kernel_specific file for hppa 2.6, but that file is gone now that 2.4 is gone, and we could trim this down now. I don't know if it would get the ramdisk under the 16 mb it seems to need to be. My a500 is insufficiently stable to do d-i development on. -- see shy jo
Attachment:
signature.asc
Description: Digital signature