"H. Peter Anvin" <hpa at zytor.com> writes: > On 11/17/2012 08:47 PM, Eric W. Biederman wrote: >>> >>> I don't think that is feasible; we need access to this structure >>> extremely early. I think if you're using the 64-bit entry point it >>> has to be an entry condition that those structures are mapped. It >>> simply becomes a precondition for using the 64-bit entry point. >> >> 64bit purgatory coming from kexec should be running with a page table >> that identity maps everything loaded by kexec and in practice all of >> memory. >> >> We should still be using that page table when we enter the new kernel. >> >> However then the 64bit kernel switches to it's own page tables. >> >> So after that point we potentially need early_ioremap, because the >> kernel's page tables may not have covered the boot_params. Certainly we >> need to do something to ensure the kernels page table covers boot_params. >> > > The easy way to fix that is to do what we do on 32 bits: we copy it > into the bss before we turn paging on (the 64-bit equivalent is before > we switch to the new page tables.) That way we know where it is, at > least for the bounded-size data items. Whatever works. It sounds like six of one half a dozen of the other to me. Last I looked early_ioremap works as soon as paging is enabled on x86_64. Eric
