Re: AMD microcode loading broken on 32 bit

[Boris Ostrovsky]

On 01/30/2014 10:13 AM, Borislav Petkov wrote:
> On Wed, Jan 29, 2014 at 12:22:19AM +0100, Borislav Petkov wrote:
> > > > Are you sure that
> > > >
> > > >     container = (u8 *)(__va((u32)relocated_ramdisk) +
> > > >                            ((u32)container -
> > > > boot_params.hdr.ramdisk_image));
> > > >
> > > > in save_microcode_in_initrd_amd() always results in a valid
> > > > pointer? It is non-NULL but it
> > > > points to address that looks to be not mapped.
> Well, here's what I see here:
>
> * initrd *without* microcode:
>
> [    0.000000] RAMDISK: [mem 0x7fbfc000-0x7ffeffff]
> [    0.000000] Allocated new RAMDISK: [mem 0x3680a000-0x36bfd099]
> [    0.000000] Move RAMDISK from [mem 0x7fbfc000-0x7ffef099] to [mem 0x3680a000-0x36bfd099]
>
> ...
>
> [    0.710771] save_microcode_in_initrd_amd:
>
> relocated_ramdisk:			0x3680a000,
> __va(relocated_ramdisk):		0xf680a000,
> container:				0x76c0e000,
> boot_params.hdr.ramdisk_image:		0x7fbfc000
>
> ...
>
> [    0.717080] BUG: unable to handle kernel paging request at 76c0e008
>
>
> * initrd + microcode:
>
> [    0.000000] RAMDISK: [mem 0x7fbf7000-0x7ffeffff]
> [    0.000000] Allocated new RAMDISK: [mem 0x36805000-0x36bfd499]
> [    0.000000] Move RAMDISK from [mem 0x7fbf7000-0x7ffef499] to [mem 0x36805000-0x36bfd499]
>
> [    0.781948] save_microcode_in_initrd_amd:
>
> relocated_ramdisk:			0x36805000,
> __va(relocated_ramdisk):		0xf6805000,
> container:				0xf680527c,
> boot_params.hdr.ramdisk_image:		0x7fbf7000
>
> Now guess what happens arithmetically if container is 0 (first case
> above without microcode):
>
> 0xf680a000 + 0 - 0x7fbfc000 = 0x76c0e000.
>
> So the fix of looking at ucode_cpio.data is correct.

That's not sufficient though, you'll need the other piece in 
find_ucode_in_initrd() where you check the header.

> An equivalent fix would be this hunk below but the initial version I
> sent you aligns more with what we do on 64-bit.
>
> ---
> diff --git a/arch/x86/kernel/cpu/microcode/amd_early.c b/arch/x86/kernel/cpu/microcode/amd_early.c
> index 8384c0fa206f..7199953fdbf2 100644
> --- a/arch/x86/kernel/cpu/microcode/amd_early.c
> +++ b/arch/x86/kernel/cpu/microcode/amd_early.c
> @@ -347,6 +347,9 @@ int __init save_microcode_in_initrd_amd(void)
>          if (!uci->cpu_sig.sig)
>                  smp_call_function_single(bsp, collect_cpu_sig_on_bsp, NULL, 1);
>   
> +       if (!container)
> +               return -EINVAL;
> +
>          /*
>           * Take into account the fact that the ramdisk might get relocated
>           * and therefore we need to recompute the container's position in

I like this one better. You can move this check above "#ifdef 
CONFIG_X86_32" and have it as the first statement in the routine, common 
between 32- and 64-bit.

-boris
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/