[U-Boot] [PATCH 12/13] armv7: support rk3066 early back to bootrom in start.S

Fri Aug 4 16:51:36 UTC 2017

> On 04 Aug 2017, at 18:33, Paweł Jarosz <paweljarosz3691 at gmail.com> wrote:
> 
> Hi,
> 
> 
> W dniu 15.06.2017 o 18:40, Simon Glass pisze:
>> Hi Pawel,
>> 
>> On 15 June 2017 at 10:32, Paweł Jarosz <paweljarosz3691 at gmail.com> wrote:
>>> 
>>> W dniu 15.06.2017 o 18:00, Simon Glass pisze:
>>> 
>>>> (just repeating other thread for completeness)
>>>> 
>>>> On 15 June 2017 at 09:42, Paweł Jarosz <paweljarosz3691 at gmail.com> wrote:
>>>>> 
>>>>> W dniu 15.06.2017 o 16:50, Simon Glass pisze:
>>>>> 
>>>>>> + Some other rockchip people
>>>>>> 
>>>>>> Hi Pawel,
>>>>>> 
>>>>>> On 15 June 2017 at 01:15, Paweł Jarosz <paweljarosz3691 at gmail.com>
>>>>>> wrote:
>>>>>>> Hi Simon
>>>>>>> 
>>>>>>> 
>>>>>>> W dniu 14.06.2017 o 13:06, Simon Glass pisze:
>>>>>>> 
>>>>>>> +Philippe
>>>>>>> 
>>>>>>> Hi Pawel,
>>>>>>> 
>>>>>>> On 12 June 2017 at 17:50, Simon Glass <sjg at chromium.org> wrote:
>>>>>>> 
>>>>>>> Hi Pawel,
>>>>>>> 
>>>>>>> On 9 June 2017 at 06:31, Paweł Jarosz <paweljarosz3691 at gmail.com>
>>>>>>> wrote:
>>>>>>> 
>>>>>>> W dniu 09.06.2017 o 13:46, Heiko Stuebner pisze:
>>>>>>> 
>>>>>>> Am Mittwoch, 7. Juni 2017, 17:37:13 CEST schrieb Paweł Jarosz:
>>>>>>> 
>>>>>>> Hi Simon,
>>>>>>> 
>>>>>>> 
>>>>>>> W dniu 06.06.2017 o 23:10, Simon Glass pisze:
>>>>>>> 
>>>>>>> Hi Pawel,
>>>>>>> 
>>>>>>> On 6 June 2017 at 12:53, Paweł Jarosz <paweljarosz3691 at gmail.com>
>>>>>>> wrote:
>>>>>>> 
>>>>>>> Rockchip bootrom first reads 1KB data from nand at offset 0x10080C00
>>>>>>> and
>>>>>>> executes it. Then waits for back to bootrom and loads another 32KB to
>>>>>>> sram
>>>>>>> which also executes. Sdram initialisation code needs to be in one of
>>>>>>> these two
>>>>>>> steps. Then bootloader loads another ~200KB of data at offset
>>>>>>> 0x60000000
>>>>>>> and jumps to it.
>>>>>>> 
>>>>>>> 32KB of data is a little low for tpl + spl part and ~200KB data is to
>>>>>>> low for
>>>>>>> u-boot part(for example to boot from mmc you need to disable usb
>>>>>>> support.
>>>>>>> 
>>>>>>> My solution to size problem is to move sdram initialisation code to tpl
>>>>>>> stage,
>>>>>>> move spl part to third stage(reading 200KB data) and add support for
>>>>>>> loading
>>>>>>> u-boot by spl from ext2/4, fat partitions.
>>>>>>> 
>>>>>>> But moving sdram initialisation code to tpl increases size of tpl above
>>>>>>> 1KB
>>>>>>> (first boot stage). Solution to this is to add code which will be below
>>>>>>> 1KB
>>>>>>> offset in tpl binary and do back to bootrom at very beginning of the
>>>>>>> tpl
>>>>>>> execution.
>>>>>>> 
>>>>>>> So do you mean that TPL starts and then loads more of itself? Why not
>>>>>>> put SDRAM init in SPL? You say above that 32KB is 'too low', but It's
>>>>>>> not clear why.
>>>>>>> 
>>>>>>> Ad.1 No. Tpl starts and at the first execution returns to bootrom.
>>>>>>> Bootrom then loads
>>>>>>> rest of the tpl (31KB) and executes it for a second time.
>>>>>>> 
>>>>>>> Ad.2,3 Due to size issues (200KB limit) i needed to move main u-boot to
>>>>>>> mmc. To load u-boot from
>>>>>>> mmc by SPL (there is 32KB bootrom limit, not enough space for mmc
>>>>>>> support) i moved SPL to sdram.
>>>>>>> Code executed in sdram can't mess with sdram settings because it will
>>>>>>> hang the board. Sdram setup
>>>>>>> needs to be done by code in SRAM (tpl).
>>>>>>> 
>>>>>>> At least the rk3288-Firefly was able to also have mmc stack in the SPL
>>>>>>> in
>>>>>>> the past, without requiring the back_to_bootrom at all. So question
>>>>>>> would
>>>>>>> be why this doesn't fit anymore, or on the rk3066 specifically.
>>>>>>> 
>>>>>>> Also, it seems like I got my hands on a preliminary (linux/mtd) nand
>>>>>>> driver
>>>>>>> (rk3228 but cursory glance at the registers suggests that it may
>>>>>>> actually
>>>>>>> work on previous socs down to the rk3066 as well) and it may be
>>>>>>> possible
>>>>>>> to adapt that for uboot, therefore making the spl able to also load the
>>>>>>> full u-boot from without needing back_to_bootrom.
>>>>>>> 
>>>>>>> 
>>>>>>> Heiko
>>>>>>> 
>>>>>>> I was not able to get mmc support on rk3066 in spl in ~31kb (32kb - 1kb
>>>>>>> for
>>>>>>> tpl)size limit.
>>>>>>> One (or two i didn't check how much) back to bootrom is required on
>>>>>>> rk3066.
>>>>>>> If not done bootrom stays in weird state and halts on bringup secondary
>>>>>>> cpu
>>>>>>> in kernel. So it's rk3066 specific.
>>>>>>> 
>>>>>>> What size do you get? With firefly-rk3288 I get about 25KB with SDRAM
>>>>>>> init and MMC stack. Are you building with Thumb 2?
>>>>>>> 
>>>>>>> If you are on irc we could try to clear this up more quicky (I am sjg1)
>>>>>>> 
>>>>>>> To summarise where I think we got to:
>>>>>>> 
>>>>>>> - move DRAM init into SPL
>>>>>>> - either have a very small TPL which just returns to boot ROM, or
>>>>>>> adjust start.S to return to the boot room early in SPL to load the
>>>>>>> other 31KB
>>>>>>> 
>>>>>>> Can you please post to the mailing list with your thoughts on this so
>>>>>>> that others (including rockchip) can chime in? I think either will
>>>>>>> work but I think others will have an opinion.
>>>>>>> 
>>>>>>> Regards,
>>>>>>> Simon
>>>>>>> 
>>>>>>> 
>>>>>>> About moving dram init to spl i agree.
>>>>>>> 
>>>>>>> I think early back to bootrom in start.S is a good solution as it would
>>>>>>> give
>>>>>>> me 1KB more space for spl and i could drop hacks like jumping to spl in
>>>>>>> tpl
>>>>>>> board file. But I would like to hear the opinion of other people on
>>>>>>> this.
>>>>>> So with this solution there would be no TPL needed? It sounds
>>>>>> reasonable to me. I'd like to hear other opinions also.
>>>>> 
>>>>> We don't need tpl if i use early back to bootrom in start.S patch with
>>>>> spl... but i didn't test it yet. If it will work i will drop the tpl. Is
>>>>> that ok?
>>>> 
>>>> It's OK with me. I don't know how the BROM re-enters SPL after the
>>>> first call into the 1KB region. Depending on how that works, it might
>>>> be too painful to use SPL only (e.g. if it jumps to a place in the
>>>> middle of SPL).
>>> 
>>> BROM re-enters same address 0x10080C04 two times. First when it boots,
>>> second after first back to brom. That is why i used counter at address
>>> 0x10080900 to count boot entries.
>>> So it's not painful at all.
>> OK from my POV this seems good.
>> 
>> [..]
>> 
>> Regards,
>> Simon
> It turns out that i will be needing tpl, spl and main u-boot. Also sdram initialization will be needed in tpl.
> 
> Reason for that is how bootrom on rk3066 executes stuff:
> 
> 1. At first stage it reads 2KB of data from nand to sram at address 0x10080C00 and decodes 1KB, executes decoded data at 0x10080C04 and waits for back to bootrom.
> 2. At second stage it reads 31KB, decodes it and executes again data at 0x10080C04 and waits for second back to bootrom. Also SDRAM init MUST BE HERE or board hangs at third stage.
> 3. At third stage bootrom reads ~200KB data to RAM at 0x60000000, decodes it and do final execution.

Are both the first stage and the second stage loaded to SRAM + 0xc00?
I.e. does the second stage (31KB) overwrite the first stage?

> All three stages are required or else later kernel will halt at bringup secondary cpus using bootrom (there is another back to bootrom in kernel code to bringup secondary cpus)
> 
> So there are two ways to boot main u-boot:
> 
> 1. TPL + SPL loaded to sram (stages one and two) and main u-boot loaded to sdram(third stage). In this case main u-boot must fit in ~200KB which is not much and ONLY ONE of the features nand, mmc, usb might be compiled in.
> 
> 2. TPL with SDRAM INIT + early back to bootrom patch as stage one and two, SPL as stage three, and main u-boot loaded from nand by SPL (stage four).
> 
> I know all of this from bootrom disassembly and tests.
> 
> So is it ok to go the second way?