"H. Peter Anvin" <hpa at zytor.com> writes: > On 12/27/2012 03:23 PM, Daniel Kiper wrote: >>> On 12/26/2012 06:18 PM, Daniel Kiper wrote: >>>> Add i386 kexec/kdump implementation. >>>> >>>> v2 - suggestions/fixes: >>>> - allocate transition page table pages below 4 GiB >>>> (suggested by Jan Beulich). >>> >>> Why? >> >> Sadly all addresses are passed via unsigned long >> variable to kexec hypercall. >> > > So can you unf*ck your broken interface before imposing it on anyone > else? Hasn't 32bit dom0 been retired? Untill the kexec firmware pass through and the normal kexec code paths are separated I can't make sense out of this code. The normal kexec code path should be doable without any special constraints on the kernel. Just leaning on some xen paravirt calls. The kexec pass through probably should not even touch x86 arch code. Certainly the same patch should not have code for both the xen kexec pass through and the paravirt arch code for the normal kexec path. Eric
