On Mon, Nov 04, 2013 at 08:53:41PM +0400, Gleb Smirnoff wrote: > On Mon, Nov 04, 2013 at 06:38:25PM +0200, Konstantin Belousov wrote: > K> Also please show me the CPU features banner from the boot in the VB, > K> like this: > > I have already collected them all. See attaches. > > Legend: > > dmesg.bb pwd_mkdb crashes during install > dmesg.milu.png pwd_mkdb crashes during install > dmesg.think installation succeeds, but world build crashes instantly > dmesg.ru all works > dmesg.morannon all works
First, all reported instances have ata attachment for the ada0, except of milu (possibly). So this means that kernel does transient remapping, and very different code path is executed comparing with what I thought initially. The path is simpler than the pure unmapped i/o. Second, I use QEMU with the ata0 attachment for the disks regularly, and I do not have an issue. I also did not see a report from the real h/w. Third point is that all reports are i386. Is there anybody with amd64, vbox, ata and the same corruption hiddent by disabling unmapped i/o ? In what way the non-working images were installed ? Is it possible to produce the problematic installs by doing it one way, and non-problematic by another ?
pgpqGG70LZE0H.pgp
Description: PGP signature
