On 13.06.2010, at 18:44, Aurelien Jarno wrote: > On Sun, Jun 13, 2010 at 09:02:40AM -0700, Richard Henderson wrote: >> On 06/13/2010 03:49 AM, Aurelien Jarno wrote: >>>> Also, what era is that second machine without highgprs? Is it running an >>>> old kernel, or a 32-bit kernel? >>> >>> I have very few infos about it, it's an IBM System z10 machine running a >>> 64-bit 2.6.26 kernel. >> >> Ah, I see it now: ea2a4d3a3a929ef494952bba57a0ef1a8a877881 >> >> [S390] 64-bit register support for 31-bit processes >> >> which adds a mechanism to pass the high parts of the gprs >> in the ucontext to the 31-bit signal handler, and adds a >> spot for them in the 31-bit core dump. >> >> It doesn't change the actual saving of registers within >> the kernel. Since we take asynchronous signals and return >> from them (as opposed to always longjmping out), we cannot >> use the full 64-bit register within a 31-bit process without >> having that bit set in HWCAP. >> >> Something to remember if we ever implement TCG for 31-bit mode. >> At the moment we only allow KVM in 31-bit mode. >> > > Is KVM in 31-bit mode actually functional?
I'm not aware of anything preventing it to be. But I honestly haven't tried. As long as all hypercall parameters stay within the first 32/31 bits, things should be safe. Alex