Re: [PATCH 08/15] efi/x86: Move command-line initrd loading to efi_main

[Dan Williams]

On Wed, May 27, 2020 at 3:47 PM Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
>
> On Wed, May 27, 2020 at 10:30:18PM +0000, Williams, Dan J wrote:
> > On Fri, 2020-05-08 at 20:01 +0200, Ard Biesheuvel wrote:
> > > From: Arvind Sankar <nivedita@xxxxxxxxxxxx>
> > >
> > > Consolidate the initrd loading in efi_main.
> > >
> > > The command line options now need to be parsed only once.
> > >
> > > Signed-off-by: Arvind Sankar <nivedita@xxxxxxxxxxxx>
> > > Link:
> > > https://lore.kernel.org/r/20200430182843.2510180-9-nivedita@xxxxxxxxxxxx
> > > Signed-off-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
> >
> > Hi,
> >
> > This patch patch in tip/master as:
> >
> > 987053a30016 efi/x86: Move command-line initrd loading to efi_main
> >
> > ...regresses my nfs root configuration. It hangs trying to mount the
> > nfs root filesystem "root=/dev/nfs ip=dhcp".
> >
> > It does not revert cleanly.
> >
> >
>
> Does this fix it?
>
> diff --git a/drivers/firmware/efi/libstub/x86-stub.c b/drivers/firmware/efi/libstub/x86-stub.c
> index defeb6035109..f53362efef84 100644
> --- a/drivers/firmware/efi/libstub/x86-stub.c
> +++ b/drivers/firmware/efi/libstub/x86-stub.c
> @@ -771,10 +771,12 @@ unsigned long efi_main(efi_handle_t handle,
>                         efi_err("Failed to load initrd!\n");
>                         goto fail;
>                 }
> -               efi_set_u64_split(addr, &hdr->ramdisk_image,
> -                                 &boot_params->ext_ramdisk_image);
> -               efi_set_u64_split(size, &hdr->ramdisk_size,
> -                                 &boot_params->ext_ramdisk_size);
> +               if (size > 0) {
> +                       efi_set_u64_split(addr, &hdr->ramdisk_image,
> +                                         &boot_params->ext_ramdisk_image);
> +                       efi_set_u64_split(size, &hdr->ramdisk_size,
> +                                         &boot_params->ext_ramdisk_size);
> +               }

I'll give it a shot, but my guess would have been something related to
the fact that this patch moves the initrd loading relative to when the
command line is being parsed. In this case it's a dracut initrd built
by:

    dracut -m "nfs network base"

...with a kernel built with:

CONFIG_IP_PNP_DHCP=y

...and a built-in network interface. The behavior seems to be that the
kernel gets an IP address just fine, but there's no initrd userspace
to mount nfs and the kernel eventually gives up looking for root.