Clara BW Fastboot / replace kernel with clara colour one possible

[matthiasmr]

Hey there,

I have a tolino shine 5 (same as kobo clara BW) and just played around with it. I came here to share my findings. They are valid from firmware 5+, but some parts seem to be there before already.

Beware: The following commands can brick your device _permanently_! Use at your own risk and own understanding. Also, you might accidentally hard reset your device deleting _all_ data!

Fastboot:
Device is fastboot enabled. Accessing is a little bit tricky:
* Turn device off and wait a moment to be sure it's off.
* Unplug the USB cable
* run your fastboot command like `fastboot oem readlkver` already, it will wait for the device
* Turn device on, keeping power button pressed
* very shortly after (within a few tenths of a sec) plug in the USB cable
* LEDS should flash very quickly.
* Keep power button pressed a bit longer, you'll see fastboot working. Feel free to release power button now.

To run a second fastboot command, unplug and replug USB cable. The device stays in fastboot mode.

With `fastboot reboot` you'll boot to bootloader which will overwrite system partition or maybe even hard reset your device!. To actually reboot, long-press the power button.

* `fastboot flash system_a path_to_img` flashes system
* `fastboot flash boot_a path_to_img` flashes kernel

You should not try to flash uboot or tee or bl2 - breaking any of those partitions renders fastboot unusable.

I went further and flashed the spa-colour kernel on my spa-bw device (via fastboot - so I knew if it flashed something I could revert by flashing the old one ).
Actually, flashing via fastboot is way quicker than using the update.tar method.

I compared the device trees of colour and bw before: They _seemed_ to be equivalent despite colour having a dual core, 2 GHz CPU (bw: single core, 1 GHz). I did not see that the touch screen controller is initialized with a different register setting - now my touch screen is inverted on the X-Axis and I didn't bother finding a way to reverse (I guess flashing back old kernel would work).

Unfortunately, my second CPU does not come up:

lscpu
Architecture:           armv7l
  Byte Order:           Little Endian
CPU(s):                 2
  On-line CPU(s) list:  0
  Off-line CPU(s) list: 1
Vendor ID:              ARM
  Model name:           Cortex-A53
    Model:              4
    Thread(s) per core: 1
    Core(s) per socket: 1
    Socket(s):          1
    Stepping:           r0p4
    CPU max MHz:        2000.0000
    CPU min MHz:        600.0000
    BogoMIPS:           15.60
    Flags:              half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm aes pmull sha1 sha2 crc32

Trying to put it online logs a kernel error message -22 (Invalid argument).

I have seen a `/proc/cpuinfo` dump of a dual core tolino vision colour which also only showed _one_ cpu core - I expect there could be a problem with the kernel that Kobo might not have noticed themselves

[matthiasmr]

I did a bit of research about the touch screen:
The difference is the CY_MT_FLAG_INV_Y Bit which invertes the Y axis before registering the touch, so it is a pure software thing, but unfortunately I cannot change this parameter after booting. So the only way to use this properly is to invert that again in KoReader or proxy the input device both seem a bit complicated...

[matthiasmr]

I could not withstand...

touch_mirrored_x = false;

in my device config in koreader frontend/device/kobo/device.lua ...

It was way to easy. Seems like I'll run with 2 GHz from now on until the next firmware update :-)

[matthiasmr]

I attached a boot log of my Kobo Clara BW bootet with the Kobo Clara Colour kernel. Maybe someone spots why the second CPU is not coming up?

[matthiasmr]

Okay, I just got confirmation that for Tolino vision color (=Kobo Libra Colour) it is the same - lscpu:

Architecture: armv7l
Byte Order: Little Endian
CPU(s): 2
On-line CPU(s) list: 0
Off-line CPU(s) list: 1
Vendor ID: ARM
Model name: Cortex-A53
Model: 4
Thread(s) per core: 1
Core(s) per socket: 1
Socket(s): 1
Stepping: r0p4
CPU max MHz: 2000.0000
CPU min MHz: 600.0000
BogoMIPS: 52.00
Flags: half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lp ae evtstrm aes pmull sha1 sha2 crc32

[matthiasmr]

Before any rumours are spread: Maybe I am wrong. On said vision colour the second core might have been turned off by koreader. I'll slow down here and verify again first. Sorry for the noise!

[DNSB]

Originally Posted by matthiasmr

I attached a boot log of my Kobo Clara BW bootet with the Kobo Clara Colour kernel. Maybe someone spots why the second CPU is not coming up?

If I remember correctly, the Clara BW has a MediaTek MT8113L single core ARM A53 1 GHz while the Clara Colour and Libra Colour have a MediaTek MT8113T dual-core ARM A53 2 GHz.

If this is correct, there is not second CPU on the Clara BW.

[matthiasmr]

While this could be true, I have following indicators to it not being true:
- bw & color have the exact same cpu core voltages for the frequencies they share - so the MT8113L is likely not another "lower power" process. So it is unlikely, that the cores are not similarly capable as the other one. I dod not see any stability problems in my tests so far, but I didn't try long.
- If you search the internet for MT8113L it likely doesn't exist
- Even if it exists, I can't believe it makes sense to save a few cents per piece and need a different stack of PCBs for that.

the ntx firmware file is different for bw & color - also there might be a hwconfig bit doing some initialisation that doesn't allow the second core to come up on my device.
Or you are just right and it doesn't exist

[matthiasmr]

So a last update on the CPU thingy:
First of all, the Tolino Vision Color / Kobo Libra Colour really has a dual core CPU. I am sorry for spreading the rumours of it not working, I was not aware that Koreader is actually shutting down the cores when idling.

I also compared the hwconfig between those and could not spot a relevant difference.

So:
* Either there is a difference in the NTX firmware binary disabling the second core on mine
* Or it _actually_ does not exist.

I stop here now. Still good to know it's possible to overclock to 2 GHz by replacing the kernel, as long as you accept the quirks with the touch screen inversion.

Some speed data: (start of display change/update finished, recorded at 30 fps)
2GHz:
Epub, Koreader,
Normal page flip: 4/9 frames, 6/11 frames 5/10 frames
going from cover page to next page: 6/11 frames
Going back to cover: 30/37 frames

1GHz:
Epub, Koreader, going from image page to next page: 7/13 frames
Going back to image: 26/31 frames
Normal page flip: 6/11 frames 5/10 frames

So... I am really curious why there was no actual speed improvement, honestly it even felt quicker with the fast CPU, but Koreader is really nice already.
Nickel @ 1 GHz, same epub:
page flip: 13/18 13/18 14/19
cover load: 22/30
cover leave: 14/19

Wow...