Boot Linux FIT image from SPL

Boot Linux FIT image from SPL

[Sven Schwermer]

Hi,

I was wondering whether it's possible to boot a Linux FIT image (kernel 
zImage, DTB, initramfs) straight from SPL? Specifically, I'm trying to 
reduce the boot time when booting into a tiny Linux with RAM disk over 
USB SDP (i.MX6ULL). This works fine with the chain ROM -> SPL -> u-boot 
proper -> Linux. I have given it a try to boot the FIT image straight 
from SPL, but I'm getting a few (debug-level) errors:

U-Boot SPL 2021.10-00003-gc254f7c98d-dirty (Dec 27 2021 - 20:15:09 +0100)
Trying to boot from USB SDP
SDP: initialize...
SDP: handle requests...
Downloading file of size 2440036 to 0x81000000... done
Jumping to header at 0x81000000
Header Tag is not an IMX image
Found header at 0x81000000
fit read sector 0, sectors=2440036, dst=88302888, count=2440036, 
size=0x253b64
cannot find property 'firmware': -1
kernel: 'kernel-1'
Embedded data: dst=82000000, size=13caf0
Image OS is Linux
fdt: 'fdt-imx6ull-myir-mys-6ulx-eval.dtb'
Embedded data: dst=8213caf0, size=64f7
cannot find property 'loadables': -1

The "loadables" config property seems to be specific to the u-boot 
proper FIT image. I have consulted the falcon-mode readme, but it 
doesn't mention FIT images which leads me to believe that this is not an 
intended use case?

Best regards,
Sven

Re: Boot Linux FIT image from SPL

[Fabio Estevam]

Hi Sven,

On Tue, Dec 28, 2021 at 8:57 AM Sven Schwermer <sven@svenschwermer.de> wrote:
>
> Hi,
>
> I was wondering whether it's possible to boot a Linux FIT image (kernel
> zImage, DTB, initramfs) straight from SPL? Specifically, I'm trying to
> reduce the boot time when booting into a tiny Linux with RAM disk over
> USB SDP (i.MX6ULL). This works fine with the chain ROM -> SPL -> u-boot
> proper -> Linux. I have given it a try to boot the FIT image straight
> from SPL, but I'm getting a few (debug-level) errors:
>
> U-Boot SPL 2021.10-00003-gc254f7c98d-dirty (Dec 27 2021 - 20:15:09 +0100)
> Trying to boot from USB SDP
> SDP: initialize...
> SDP: handle requests...
> Downloading file of size 2440036 to 0x81000000... done
> Jumping to header at 0x81000000
> Header Tag is not an IMX image
> Found header at 0x81000000
> fit read sector 0, sectors=2440036, dst=88302888, count=2440036,
> size=0x253b64
> cannot find property 'firmware': -1
> kernel: 'kernel-1'
> Embedded data: dst=82000000, size=13caf0
> Image OS is Linux
> fdt: 'fdt-imx6ull-myir-mys-6ulx-eval.dtb'
> Embedded data: dst=8213caf0, size=64f7
> cannot find property 'loadables': -1
>
> The "loadables" config property seems to be specific to the u-boot
> proper FIT image. I have consulted the falcon-mode readme, but it
> doesn't mention FIT images which leads me to believe that this is not an
> intended use case?

People load FIT images via Falcon mode on Layerscape:
https://source.denx.de/u-boot/u-boot/-/blob/master/arch/arm/cpu/armv8/fsl-layerscape/doc/README.falcon4

Hope this helps.

Regards,

Fabio Estevam

Re: Boot Linux FIT image from SPL

[Fabio Estevam]

On Tue, Dec 28, 2021 at 9:13 AM Fabio Estevam <festevam@gmail.com> wrote:

> People load FIT images via Falcon mode on Layerscape:
> https://source.denx.de/u-boot/u-boot/-/blob/master/arch/arm/cpu/armv8/fsl-layerscape/doc/README.falcon4

There was a typo in the URL above. Correct one is:

https://source.denx.de/u-boot/u-boot/-/blob/master/arch/arm/cpu/armv8/fsl-layerscape/doc/README.falcon

Re: Boot Linux FIT image from SPL

[Sven Schwermer]

Hi Fabio,

Thank you for the hint. That's a _very_ useful document. In the 
meantime, I have been able to boot my Linux FIT image straight from SPL 
on my i.MX6ULL platform.

I have, however, noticed that Linux boots much faster when launched from 
u-boot proper compared to SPL. Time to (initramfs) init is 1.2s vs. 5s. 
That is with identical FIT images (kernel, device tree, initramfs, 
kernel command line). I ruled out differing core clock rates (both run 
at 396MHz), caches (caches seem to be disabled before handing over to 
Linux) and differing load addresses (might cause additional relocations, 
copies).

What other setup could u-boot proper do that would make Linux boot 
faster that I haven't thought about?

Best regards,
Sven

On 12/28/21 13:15, Fabio Estevam wrote:
> On Tue, Dec 28, 2021 at 9:13 AM Fabio Estevam <festevam@gmail.com> wrote:
> 
>> People load FIT images via Falcon mode on Layerscape:
>> https://source.denx.de/u-boot/u-boot/-/blob/master/arch/arm/cpu/armv8/fsl-layerscape/doc/README.falcon4
> 
> There was a typo in the URL above. Correct one is:
> 
> https://source.denx.de/u-boot/u-boot/-/blob/master/arch/arm/cpu/armv8/fsl-layerscape/doc/README.falcon

Re: Boot Linux FIT image from SPL

[Fabio Estevam]

Hi Sven,

On Thu, Dec 30, 2021 at 6:22 PM Sven Schwermer <sven@svenschwermer.de> wrote:
>
> Hi Fabio,
>
> Thank you for the hint. That's a _very_ useful document. In the
> meantime, I have been able to boot my Linux FIT image straight from SPL
> on my i.MX6ULL platform.

Glad you got it to work. It would be nice if you could submit a patch
adding a document
adapted for i.MX6ULL.

> I have, however, noticed that Linux boots much faster when launched from
> u-boot proper compared to SPL. Time to (initramfs) init is 1.2s vs. 5s.
> That is with identical FIT images (kernel, device tree, initramfs,
> kernel command line). I ruled out differing core clock rates (both run
> at 396MHz), caches (caches seem to be disabled before handing over to
> Linux) and differing load addresses (might cause additional relocations,
> copies).
>
> What other setup could u-boot proper do that would make Linux boot
> faster that I haven't thought about?

Looking at your previous log, I see that you boot from serial download mode.

I recall seeing such a slow boot behavior when booting from SDP and
this was fixed by the following commit:

commit c5437e5b8aff9c952ebaab9be7670439c141e4e7
Author: Ye Li <ye.li@nxp.com>
Date:   Mon May 14 09:44:29 2018 -0300

    imx: Enable ACTLR.SMP bit for all i.MX cortex-a7 platforms

    According to the Cortex-A7 TRM, for ACTLR.SMP bit "You must ensure this bit
    is set to 1 before the caches and MMU are enabled, or any cache and TLB
    maintenance operations are performed".
    ROM sets this bit in normal boot flow, but when in serial download mode,
    it is not set.
    Here we add it in u-boot as a common flow for all i.MX cortex-a7 platforms,
    including mx7d, mx6ul/ull and mx7ulp.

    Signed-off-by: Ye Li <ye.li@nxp.com>
    [fabio: adapted to U-Boot mainline codebase and make checkpatch happy]
    Signed-off-by: Fabio Estevam <fabio.estevam@nxp.com>

Such slow boot is not observed if you boot from other media, such as
eMMC, for example.

You could try performing the setting of ACTLR.SMP bit inside SPL.
Currently, it is done only inside U-Boot proper.

Hope this helps.

Re: Boot Linux FIT image from SPL

[Sven Schwermer]

Hi Fabio,

> Glad you got it to work. It would be nice if you could submit a patch
> adding a document adapted for i.MX6ULL.

I'll need to sort my notes and go through my changes. I'll try to submit 
relevant patches.

> You could try performing the setting of ACTLR.SMP bit inside SPL.
> Currently, it is done only inside U-Boot proper.
> 
> Hope this helps.

This does indeed make all the difference. Thank you very much for the 
tip! I'll send a patch soon.

Best regards,
Sven

Re: Boot Linux FIT image from SPL

[Sven Schwermer]

Hi Fabio,

> You could try performing the setting of ACTLR.SMP bit inside SPL.
> Currently, it is done only inside U-Boot proper.

I have one more question about this: Is it save to just enable D/I 
caches (enable_caches) in SPL for all of mach-imx or will this have 
unforeseen side effects? In that case, I can just make enable_ca7_smp 
non-static and call that for all of mach-imx in SPL. In either case, 
what would be the proper spot for this? I'm thinking somewhere in SPL's 
board_init_r or spl_init/spl_common_init?

Best regards,
Sven

Re: Boot Linux FIT image from SPL

[Fabio Estevam]

Hi Sven,

On Fri, Dec 31, 2021 at 11:38 AM Sven Schwermer <sven@svenschwermer.de> wrote:

> I have one more question about this: Is it save to just enable D/I
> caches (enable_caches) in SPL for all of mach-imx or will this have
> unforeseen side effects? In that case, I can just make enable_ca7_smp
> non-static and call that for all of mach-imx in SPL. In either case,
> what would be the proper spot for this? I'm thinking somewhere in SPL's
> board_init_r or spl_init/spl_common_init?

To be on the safe side, I would only call enable_ca7_smp() for SPL.

What about calling it from arch_cpu_init() in arch/arm/mach-imx/mx6/soc.c?

Regards,

Fabio Estevam



>
> Best regards,
> Sven