[PATCH] sunxi-common: support loading compressed kernel images on ARMv8

[Samuel Holland]

Andre,

On 4/17/21 1:23 PM, Andre Przywara wrote:
> On Sat, 17 Apr 2021 12:47:23 -0500 Samuel Holland <samuel at sholland.org> wrote:
>> On 3/31/21 10:39 AM, Andre Przywara wrote:
>>> On 20/02/2021 12:14, Nicolas Boulenguez wrote:
>>>
>>> Hi Arnaud,
>>>   
>>>> From: Arnaud Ferraris <arnaud.ferraris at gmail.com>  
>>>
>>> as mentioned before, this looks useful, so can you re-send this with a
>>> commit message and your Signed-off-by:?  
>>
>> It looks like you picked up this patch regardless?
> 
> Yes, I asked him twice, without any reply. At the end I figured that
> the danger of a copyright infringement is low given the nature of the
> change, and I didn't feel this should block this useful feature. The
> alternative would be to have sent it as a patch of mine, but I felt
> that I should preserve his authorship, so took the patch anyway.
> I got a reply from him on another patch, and it's not his first U-Boot
> patch, so I think it's just an oversight (from the sender of the patch?)
> 
> Chen-Yu asked about the feature lately, so if you think that approach
> is dodgy, we can revert it and take another version of the patch.

Okay, I'll trust your judgement here.

>>> Also I was wondering if you could increase the COMP_SIZE? The 64MB look
>>> rather arbitrary and maybe unnecessarily limiting. I think we could go
>>> up till 0xfa00000, but maybe should stop at 0xf800000, to leave some
>>> space for further extensions?  
>>
>> And you changed KERNEL_COMP_SIZE to 0xb000000. How does that work when
>> U-Boot itself is at 0x4a000000 (KERNEL_COMP_ADDR_R + 0x6000000)?
> 
> It's not anymore at this time, that's just the load address, and it
> relocates itself to the end of DRAM very early. So basically the whole
> memory except the very end is available.
>
> Btw, the reason we confine everything to below 0x50000000 is just that
> we need to concatenate the base (0x4....) with that offset, and cannot
> use arithmetic, because that doesn't work in the Makefiles.

Ah, that makes sense. So we're assuming that U-Boot relocates itself
above 0x50000000, and so nothing important remains inside the address
range used by sunxi-common.h.

Thanks for the explanation,
Samuel

> Cheers,
> Andre
> 
>>>> ---
>>>>  include/configs/sunxi-common.h | 15 +++++++++++++++
>>>>  1 file changed, 15 insertions(+)
>>>>
>>>> diff --git a/include/configs/sunxi-common.h b/include/configs/sunxi-common.h
>>>> index 5b0bec0561..227284cd29 100644
>>>> --- a/include/configs/sunxi-common.h
>>>> +++ b/include/configs/sunxi-common.h
>>>> @@ -291,6 +291,8 @@ extern int soft_i2c_gpio_scl;
>>>>   */
>>>>  #define BOOTM_SIZE	__stringify(0xa000000)
>>>>  #define KERNEL_ADDR_R	__stringify(SDRAM_OFFSET(0080000))
>>>> +#define KERNEL_COMP_ADDR_R	__stringify(SDRAM_OFFSET(4000000))
>>>> +#define KERNEL_COMP_SIZE	__stringify(0x4000000)
>>>>  #define FDT_ADDR_R	__stringify(SDRAM_OFFSET(FA00000))
>>>>  #define SCRIPT_ADDR_R	__stringify(SDRAM_OFFSET(FC00000))
>>>>  #define PXEFILE_ADDR_R	__stringify(SDRAM_OFFSET(FD00000))
>>>> @@ -332,6 +334,18 @@ extern int soft_i2c_gpio_scl;
>>>>  	"pxefile_addr_r=" PXEFILE_ADDR_R "\0" \
>>>>  	"ramdisk_addr_r=" RAMDISK_ADDR_R "\0"
>>>>  
>>>> +#ifdef CONFIG_ARM64
>>>> +
>>>> +#define MEM_LAYOUT_ENV_EXTRA_SETTINGS \
>>>> +	"kernel_comp_addr_r=" KERNEL_COMP_ADDR_R "\0" \
>>>> +	"kernel_comp_size=" KERNEL_COMP_SIZE "\0"
>>>> +
>>>> +#else
>>>> +
>>>> +#define MEM_LAYOUT_ENV_EXTRA_SETTINGS ""  
>>
>> This can be defined to nothing. No empty string is needed.
>>
>>>> +
>>>> +#endif
>>>> +
>>>>  #define DFU_ALT_INFO_RAM \
>>>>  	"dfu_alt_info_ram=" \
>>>>  	"kernel ram " KERNEL_ADDR_R " 0x1000000;" \
>>>> @@ -487,6 +501,7 @@ extern int soft_i2c_gpio_scl;
>>>>  #define CONFIG_EXTRA_ENV_SETTINGS \
>>>>  	CONSOLE_ENV_SETTINGS \
>>>>  	MEM_LAYOUT_ENV_SETTINGS \
>>>> +	MEM_LAYOUT_ENV_EXTRA_SETTINGS \
>>>>  	DFU_ALT_INFO_RAM \
>>>>  	"fdtfile=" FDTFILE "\0" \
>>>>  	"console=ttyS0,115200\0" \
>>>>  
>>>   
>>
>