[PATCH v2 1/6] efi_loader: add RAM disk device path

[Masahisa Kojima]

On Tue, 18 Jul 2023 at 14:57, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
>
>
>
> Am 18. Juli 2023 03:31:30 MESZ schrieb Masahisa Kojima <masahisa.kojima at linaro.org>:
> >Hi Heinrich,
> >
> >On Fri, 14 Jul 2023 at 22:44, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
> >>
> >> On 14.07.23 07:44, Masahisa Kojima wrote:
> >> > This is a preparation to add the EFI_RAM_DISK_PROTOCOL.
> >> > This commit adds the RAM disk device path structure
> >> > and text conversion to Device Path to Text Protocol.
> >> >
> >> > Signed-off-by: Masahisa Kojima <masahisa.kojima at linaro.org>
> >> > ---
> >> > No update since v1
> >> >
> >> >   include/efi_api.h                        | 19 +++++++++++++++++++
> >> >   lib/efi_loader/efi_device_path_to_text.c | 14 ++++++++++++++
> >> >   2 files changed, 33 insertions(+)
> >> >
> >> > diff --git a/include/efi_api.h b/include/efi_api.h
> >> > index 55a4c989fc..4ee4a1b5e9 100644
> >> > --- a/include/efi_api.h
> >> > +++ b/include/efi_api.h
> >> > @@ -682,6 +682,7 @@ struct efi_device_path_uri {
> >> >   #  define DEVICE_PATH_SUB_TYPE_CDROM_PATH   0x02
> >> >   #  define DEVICE_PATH_SUB_TYPE_VENDOR_PATH  0x03
> >> >   #  define DEVICE_PATH_SUB_TYPE_FILE_PATH    0x04
> >> > +#  define DEVICE_PATH_SUB_TYPE_RAM_DISK_PATH 0x09
> >> >
> >> >   struct efi_device_path_hard_drive_path {
> >> >       struct efi_device_path dp;
> >> > @@ -705,6 +706,24 @@ struct efi_device_path_file_path {
> >> >       u16 str[];
> >> >   } __packed;
> >> >
> >> > +/* This GUID defines a RAM Disk supporting a raw disk format in volatile memory */
> >> > +#define EFI_VIRTUAL_DISK_GUID \
> >> > +     EFI_GUID(0x77ab535a, 0x45fc, 0x624b, \
> >> > +     0x55, 0x60, 0xf7, 0xb2, 0x81, 0xd1, 0xf9, 0x6e)
> >> > +
> >> > +/* This GUID defines a RAM Disk supporting an ISO image in volatile memory */
> >> > +#define EFI_VIRTUAL_CD_GUID \
> >> > +     EFI_GUID(0x3d5abd30, 0x4175, 0x87ce, \
> >> > +              0x6d, 0x64, 0xd2, 0xad, 0xe5, 0x23, 0xc4, 0xbb)
> >> > +
> >> > +struct efi_device_path_ram_disk_path {
> >> > +     struct efi_device_path dp;
> >> > +     u64 starting_address;
> >> > +     u64 ending_address;
> >> > +     efi_guid_t disk_type_guid;
> >> > +     u16 disk_instance;
> >> > +} __packed;
> >> > +
> >> >   #define EFI_BLOCK_IO_PROTOCOL_GUID \
> >> >       EFI_GUID(0x964e5b21, 0x6459, 0x11d2, \
> >> >                0x8e, 0x39, 0x00, 0xa0, 0xc9, 0x69, 0x72, 0x3b)
> >> > diff --git a/lib/efi_loader/efi_device_path_to_text.c b/lib/efi_loader/efi_device_path_to_text.c
> >> > index 8c76d8be60..4395e79f33 100644
> >> > --- a/lib/efi_loader/efi_device_path_to_text.c
> >> > +++ b/lib/efi_loader/efi_device_path_to_text.c
> >> > @@ -324,6 +324,20 @@ static char *dp_media(char *s, struct efi_device_path *dp)
> >> >               free(buffer);
> >> >               break;
> >> >       }
> >> > +     case DEVICE_PATH_SUB_TYPE_RAM_DISK_PATH: {
> >> > +             struct efi_device_path_ram_disk_path *rddp =
> >> > +                     (struct efi_device_path_ram_disk_path *)dp;
> >> > +             u64 start;
> >> > +             u64 end;
> >> > +
> >> > +             /* Copy from packed structure to aligned memory */
> >> > +             memcpy(&start, &rddp->starting_address, sizeof(start));
> >> > +             memcpy(&end, &rddp->ending_address, sizeof(end));
> >> > +
> >> > +             s += sprintf(s, "RamDisk(0x%llx,%llx,%pUl,0x%x)", start, end,
> >> > +                          &rddp->disk_type_guid, rddp->disk_instance);
> >>
> >> If there is no alignment guarantee for starting_address, then the same
> >> is true for disk_instance which may spread over two u64 blocks.
> >
> >disk_instance is a u16 field, so it is aligned.
>
> A preceding node, e.g. VenHw(), may have an uneven length?
>
> >
> >>
> >> In case of DEVICE_PATH_SUB_TYPE_MEMORY we don't use memcpy() to align u64.
> >>
> >> I don't think we call device_path_to_text before allow_unaligned().
> >>
> >> There is a family of functions like get_unaligned_le64() if we should
> >> ever need to a align a value. Or we could copy the whole device path node.
> >
> >OK, I will use get_unaligned_le64().
>
> Why do you think this is necessary?

Sorry, I misunderstood your comment.
allow_unaligned() is called in efi_init_early(), device_path_to_text
will not be called
before allow_unaligned(). So we don't need to care about alignment in
device_path_to_text
like DEVICE_PATH_SUB_TYPE_MEMORY is doing.

Thanks,
Masahisa Kojima

>
> Best regards
>
> Heinrich
>
> >
> >Thanks,
> >Masahisa Kojima
> >
> >>
> >> Best regards
> >>
> >> Heinrich
> >>
> >> > +             break;
> >> > +     }
> >> >       default:
> >> >               s = dp_unknown(s, dp);
> >> >               break;
> >>