On 2017-02-03 10:11, Henning Schild wrote: > On Thu, 2 Feb 2017 18:36:59 +0100 > Jan Kiszka <[email protected]> wrote: > >> On 2017-02-02 18:17, Ralf Ramsauer wrote: >>> Hi, >>> >>> upstream Linux commit 3277953 (mm: do not export ioremap_page_range >>> symbol for external module) unexports the ioremap_page_range symbol, >>> which we need for jailhouse_ioremap(). >>> >>> The unexport was introduced in 4.10-rc6. To be able to load >>> Jailhouse with Linux 4.10-rc6+, we either need to revert this >>> commit, or use another API. >>> >>> Ralf >>> >>> https://github.com/torvalds/linux/commit/3277953de2f31dd03c6375e9a9f680ac37fc9d27 >>> >> >> Yeah, this increases pressure to rethink the handover procedure /wrt >> address spaces. The simple workaround will be a kernel patch. >> Long-term, it might be better to make Jailhouse relocatable (again) >> and do this in the very early boot phase. Quite a bit of work, >> though... > > Not sure whether our case would be strong enough to get that change > reverted upstream. At least looking at the discussion that led to the > change and maybe talking to the author are probably not bad ideas. > > The comment suggest there are plenty of alternatives, did not look > further but it might just be a matter of using another ioremap_.
Currently, out requirement on the API is to define our own virtual address for the mapping. All the mentioned ioremap_* variants do not seem to cover this. However, I'd first like to understand if we will need this fixed virtual mapping on the Linux side also on the long run - before asking the kernel to do some still out-of-tree component a favour. It's rather unlikely they would do so anyway. Jan -- You received this message because you are subscribed to the Google Groups "Jailhouse" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
