On Sun, 2008-10-05 at 11:04 -0700, Yinghai Lu wrote:
quoted text > On Sun, Oct 5, 2008 at 1:56 AM, huang ying <huang.ying.caritas@gmail.com>=
wrote:
quoted text > > On Sun, Oct 5, 2008 at 1:44 AM, Yinghai Lu <yinghai@kernel.org> wrote:
> >> On Sat, Oct 4, 2008 at 2:35 AM, huang ying <huang.ying.caritas@gmail.c=
om> wrote:
quoted text > > [...]
> >>> Using __va and efi_ioremap() here is to make EFI support compatible
> >>> with kexec. Because EFI provide only efi_enter_virtual_mode(), no
> >>> efi_leave_virtual_mode(), we should make EFI runtime memory area
> >>> mapped to same virtual memory area in original kernel and kexeced
> >>> kernel, so that the EFI runtime services can be used in kexeced
> >>> kernel.
> >>
> >> so need to make efi range all under direct-mapping like E820-RAM?
> >
> > Some EFI runtime range is just some RAM area used by EFI runtime
> > services, they can be direct-mapped. Some EFI runtime range may be IO
> > MEM range used by EFI runtime services, it is possible that these IO
> > MEM range can not be direct-mapped. So I implement efi_ioremap() to
> > deal with them.
> >
>=20
> i'm confused.
>=20
> so
> --- linux-2.6.orig/arch/x86/kernel/efi.c
> +++ linux-2.6/arch/x86/kernel/efi.c
> @@ -475,10 +475,7 @@ void __init efi_enter_virtual_mode(void)
> size =3D md->num_pages << EFI_PAGE_SHIFT;
> end =3D md->phys_addr + size;
>=20
> - if (PFN_UP(end) <=3D max_low_pfn_mapped)
> - va =3D __va(md->phys_addr);
> - else
> - va =3D efi_ioremap(md->phys_addr, size);
> + va =3D efi_ioremap(md->phys_addr, size);
>=20
> md->virt_addr =3D (u64) (unsigned long) va;
> should be ok.
I think __va() and efi_ioremap() should be used.
quoted text > then how about use ioremap directly to replace fixed mapping in 64bit
> with efi_ioremap?
Because for kexec to work, EFI runtime memory area should be mapped to
same virtual address across reboot.
Best Regards,
Huang Ying
