Hi Jonas,
On 5/21/24 1:57 PM, Jonas Karlman wrote:
Hi Quentin,
On 2024-05-21 11:53, Quentin Schulz wrote:
> Hi all,
>
> On 5/20/24 6:06 PM, Tom Rini wrote:
>> Hey all,
>>
>> It's release day and here is -rc3. At this point, I think we're firmly
>> on "bug fix and documentation" ground moving forward for changes to
>> v2024.07 and the next branch is open and ready for use and pull
>> requests.
>>
>> In terms of a changelog,
>> git log --merges v2024.07-rc2..v2024.07-rc3
>> contains what I've pulled but as always, better PR messages and tags
>> will provide better results here.
>>
>> I hope to remain on schedule and that means the rest of the rcs every
>> other Monday, and with final release on Monday, July 1st, 2024. Thanks
>> all!
>>
>
> Just to let you know that Rockchip PX30 support is broken since
>
https://source.denx.de/u-boot/u-boot/-/commit/1d01440b3f14f6648cd672da619489f635cb209c
> (at the very least on Ringneck, haven't tested on PX30 Mini-EVB yet).
>
> U-Boot proper pre-reloc seems to be running out of memory and we cannot
> reach U-Boot CLI anymore.
>
> I hope I'll have time to look into this before v2024.07 is released but
> cannot promise anything.
On Rockchip platform all DRAM is available at U-Boot proper pre-reloc,
so there is no need to limit SYS_MALLOC_F_LEN to a small amount.
Ringneck seem to use 8 KiB for pre-reloc malloc() pool, please try with
a 64 KiB pool, same as rk3308/rk3328/rk3399/rk356x/rk3588 all now use.
CONFIG_SYS_MALLOC_F_LEN=0x1
This made it work, thanks for the hint.
malloc() pool size used by TPL/SPL should still work:
CONFIG_TPL_SYS_MALLOC_F_LEN=0x600
CONFIG_SPL_SYS_MALLOC_F_LEN=0x2000
Also check the "Early malloc usage" from bdinfo cmd, should show how
much memory was actually allocated at pre-reloc, from a rk3308 board:
Early malloc usage: 2060 / 1
"""
=> bdinfo
boot_params = 0x
DRAM bank = 0x
-> start= 0x0020
-> size = 0x7fe0
flashstart = 0x
flashsize = 0x
flashoffset = 0x
baudrate= 115200 bps
relocaddr = 0x7ff53000
reloc off = 0x7fd53000
Build = 64-bit
current eth = ethernet@ff36
ethaddr = f6:3c:b5:0d:0a:13
IP addr =
fdt_blob= 0x7df43bd0
new_fdt = 0x7df43bd0
fdt_size= 0xc1e0
lmb_dump_all:
memory.cnt = 0x1 / max = 0x10
memory[0] [0x20-0x7fff], 0x7fe0 bytes flags: 0
reserved.cnt = 0x1 / max = 0x10
reserved[0] [0x7df3f580-0x7fff], 0x020c0a80 bytes flags: 0
devicetree = separate
serial addr = 0xff03
width = 0x0004
shift = 0x0002
offset = 0x
clock = 0x016e3600
arch_number = 0x
TLB addr= 0x7fff
irq_sp = 0x7df43bc0
sp start= 0x7df43bc0
Early malloc usage: 2250 / 1
"""
The default on PX30 for SYS_MALLOC_F_LEN is 0x400 (both from
arch/arm/mach-rockchip/px30/Kconfig and Kconfig).
The default for SPL_SYS_MALLOC_F_LEN is SYS_MALLOC_F_LEN.
The default for TPL_SYS_MALLOC_F_LEN is SPL_SYS_MALLOC_F_LEN.
I suggest we change the default SYS_MALLOC_F_LEN to be 0x2 then, on PX30.
Since we have forced TPL support on PX30, we can keep SPL_SYS_MALLOC_F_LEN is
SYS_MALLOC_F_LEN, because we necessarily have DRAM support in SPL so we are
much less limited in allocation.
Since all PX30 boards have 0x600 for TPL_SYS_MALLOC_F_LEN I suggest to make
this the default as well?
Anything against this?
I'm also a bit confused as to why we need to specify the size of the allocation
pool pre-relocation? Why is this important?
Thanks a lot for the hint, I can now work on migrating PX30 to OF_UPSTREAM :)
Cheers,
Quentin
