[PATCH] xload: be more flexible when searching for second stage bootloader.

[vj]

On Wed, Sep 30, 2015 at 8:19 AM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
> On Tue, Sep 29, 2015 at 08:06:35PM +0100, vj wrote:
>> On Tue, Sep 29, 2015 at 8:05 AM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
>> > On Sun, Sep 27, 2015 at 03:08:51PM +0100, Vicente Bergas wrote:
>> >> A first stage bootloader can read fat and ext4 filesystems, and
>> >> even both can be compiled-in at the same time.
>> >> But then xload has a hardcoded fat filesystem mount option which
>> >> renders ext4 unusable.
>> >> This patch tries to mount it as ext4 if the fat attempt fails.
>> >>
>> >> Then, a typical use case of an ext4 formatted filesystem
>> >> is it to be a standard linux filesystem, which contains
>> >> boot-related files in /boot. So, when searching for the second stage
>> >> bootloader, try /boot/barebox.bin after not finding it in /barebox.bin
>> >>
>> >> Signed-off-by: Vicente Bergas <vicencb at gmail.com>
>> >> ---
>> >>  arch/arm/mach-omap/xload.c | 4 ++++
>> >>  1 file changed, 4 insertions(+)
>> >>
>> >> diff --git a/arch/arm/mach-omap/xload.c b/arch/arm/mach-omap/xload.c
>> >> index ebcbcbc..c5e29f8 100644
>> >> --- a/arch/arm/mach-omap/xload.c
>> >> +++ b/arch/arm/mach-omap/xload.c
>> >> @@ -118,6 +118,8 @@ static void *omap_xload_boot_mmc(void)
>> >>       partname = asprintf("%s.0", diskdev);
>> >>
>> >>       ret = mount(partname, "fat", "/", NULL);
>> >> +     if (ret)
>> >> +             ret = mount(partname, "ext4", "/", NULL);
>> >
>> > It should also be possible to pass NULL as fs type in which case we
>> > automatically detect the fs type. Could you check that?
>>
>> I didn't know that. This way it would be nicer and cleaner!
>> So it has been tested but it didn't work :(
>> mount calls detect_fs, detect_fs calls file_name_detect_type and
>> file_name_detect_type calls open.
>> open fails to open the "mmc0.0" file name
>
> Ok, I see. This indeed can't work as it depends on devfs being mounted
> on /dev/. For this to work we would need a cdev_detect_filetype(const
> char *name) which uses the cdev_* operations to read a buffer and pass
> that to file_detect_type(). You're invited to implement that ;)
>
>> So, the first 128KB block can hold an MBR plus partition table, then a
>> raw image of the MLO file can be present at offset 128KB followed by
>> the disk partitions.
>>
>> In section 26.1.7.5.3 page 4939 figure 26-17 MMC/SD Booting
>> states that raw mode detection is performed before searching for the
>> MLO file in a filesystem.
>>
>> All in all, it is possible to have a "fat-free" SD-card.
>
> Now that you say it I remember having read about this before.
>
> Does this work well with the rest of barebox? In omap_env_init() we
> mount the first mmc partition to /boot and put the environment to
> /boot/barebox.env. That doesn't work when the first partition doesn't
> contain a FAT. Do you put your environment somewhere else?

I was not using the environment save/restore feature and was unaware
that it was broken.
The environment retrieval could be improved in a similar way: try
mounting all supported filesystems and search in "mountpoint/" and
"mountpoint/boot/"
But the environment save feature can't be easily supported on ext4
because the driver does not support writes.
Still I think that this patch is an improvement, is not the patch
itself that broke the env save/restore, it was a new use case enabled
by it.

>
> Sascha
>
> --
> Pengutronix e.K.                           |                             |
> Industrial Linux Solutions                 | http://www.pengutronix.de/  |
> Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
> Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |