[PATCH v4 5/9] xtensa: Remove arch_setup_bdinfo()

[Ovidiu Panait]

Hi Stefan,

On 13.08.2020 11:15, Stefan Roese wrote:
> Hi Ovidiu,
>
> On 13.08.20 10:09, Stefan Roese wrote:
>> Hi Ovidiu,
>>
>> On 13.08.20 09:57, Ovidiu Panait wrote:
>>> Hi Stefan,
>>>
>>> On 13.08.2020 08:47, Stefan Roese wrote:
>>>> arch_setup_bdinfo() only configures the deprecated bi_memstart &
>>>> bi_memsize values, which should not be needed any more. Lets remove
>>>> this file completely.
>>>>
>>>> Signed-off-by: Stefan Roese <sr at denx.de>
>>>>
>>>> ---
>>>>
>>>> Changes in v4:
>>>> - New patch
>>>>
>>>>   arch/xtensa/lib/Makefile |  2 +-
>>>>   arch/xtensa/lib/bdinfo.c | 22 ----------------------
>>>>   2 files changed, 1 insertion(+), 23 deletions(-)
>>>>   delete mode 100644 arch/xtensa/lib/bdinfo.c
>>>>
>>>> diff --git a/arch/xtensa/lib/Makefile b/arch/xtensa/lib/Makefile
>>>> index ceee59b9bd..c59df7d372 100644
>>>> --- a/arch/xtensa/lib/Makefile
>>>> +++ b/arch/xtensa/lib/Makefile
>>>> @@ -5,4 +5,4 @@
>>>>   obj-$(CONFIG_CMD_BOOTM) += bootm.o
>>>> -obj-y     += cache.o misc.o relocate.o time.o bdinfo.o
>>>> +obj-y     += cache.o misc.o relocate.o time.o
>>>> diff --git a/arch/xtensa/lib/bdinfo.c b/arch/xtensa/lib/bdinfo.c
>>>> deleted file mode 100644
>>>> index 4ec8529521..0000000000
>>>> --- a/arch/xtensa/lib/bdinfo.c
>>>> +++ /dev/null
>>>> @@ -1,22 +0,0 @@
>>>> -// SPDX-License-Identifier: GPL-2.0+
>>>> -/*
>>>> - * XTENSA-specific information for the 'bd' command
>>>> - *
>>>> - * (C) Copyright 2003
>>>> - * Wolfgang Denk, DENX Software Engineering, wd at denx.de.
>>>> - */
>>>> -
>>>> -#include <common.h>
>>>> -#include <init.h>
>>>> -
>>>> -DECLARE_GLOBAL_DATA_PTR;
>>>> -
>>>> -int arch_setup_bdinfo(void)
>>>> -{
>>>> -    struct bd_info *bd = gd->bd;
>>>> -
>>>> -    bd->bi_memstart = PHYSADDR(CONFIG_SYS_SDRAM_BASE);
>>>> -    bd->bi_memsize = CONFIG_SYS_SDRAM_SIZE;
>>>> -
>>>
>>> When I did the setup_bdinfo refactoring, I realized that xtensa was 
>>> the only arch handling bi_{memstart,memsize} in a special way:
>>>
>>>      bd->bi_memstart = PHYSADDR(CONFIG_SYS_SDRAM_BASE);
>>>      bd->bi_memsize = CONFIG_SYS_SDRAM_SIZE;
>>>
>>> Because xtensa uses PHYSADDR(CONFIG_SYS_SDRAM_BASE) for bi_memstart, 
>>> I was not able to replace it with gd->ram_base, as for all the other 
>>> arches.
>>
>> Please note that bi_memstart/size is probably not used at all on
>> xtensa - only for the bdinfo cmd AFAICT.
>>
>>> Currently, gd->ram_base is defined in common/board_f.c, function 
>>> setup_dest_addr as:
>>>
>>> #ifdef CONFIG_SYS_SDRAM_BASE
>>>> -------gd->ram_base = CONFIG_SYS_SDRAM_BASE;
>>> #endif
>>
>> Yes, this #ifdef is ugly - I also noticed it while working on this
>> patchset.
>>
>>> So I think the PHYSADDR() logic needs to be preserved so that the 
>>> patchset would not change the memory start/end logic in 
>>> arch/xtensa/lib/bootm.c:
>>>
>>> diff --git a/arch/xtensa/lib/bootm.c b/arch/xtensa/lib/bootm.c
>>> index 458eaf95c0..6fcbfc4292 100644
>>> --- a/arch/xtensa/lib/bootm.c
>>> +++ b/arch/xtensa/lib/bootm.c
>>> @@ -41,15 +41,14 @@ static struct bp_tag *setup_last_tag(struct 
>>> bp_tag *params)
>>>
>>>   static struct bp_tag *setup_memory_tag(struct bp_tag *params)
>>>   {
>>> -    struct bd_info *bd = gd->bd;
>>>       struct meminfo *mem;
>>>
>>>       params->id = BP_TAG_MEMORY;
>>>       params->size = sizeof(struct meminfo);
>>>       mem = (struct meminfo *)params->data;
>>>       mem->type = MEMORY_TYPE_CONVENTIONAL;
>>> -    mem->start = bd->bi_memstart;
>>> -    mem->end = bd->bi_memstart + bd->bi_memsize;
>>> +    mem->start = gd->ram_base;
>>> +    mem->end = gd->ram_base + gd->ram_size;
>>
>> I see. Why not add this instead as well to this new patchset:
>>
>> diff --git a/arch/xtensa/cpu/cpu.c b/arch/xtensa/cpu/cpu.c
>> index 85d3464607..a4ba71a301 100644
>> --- a/arch/xtensa/cpu/cpu.c
>> +++ b/arch/xtensa/cpu/cpu.c
>> @@ -45,6 +45,7 @@ int print_cpuinfo(void)
>>
>>   int arch_cpu_init(void)
>>   {
>> +       gd->ram_base = PHYSADDR(CONFIG_SYS_SDRAM_BASE);
>>          gd->ram_size = CONFIG_SYS_SDRAM_SIZE;
>>          return 0;
>>   }
>>
>> and keep the bi_memstart -> ram_base conversion above? Looks more
>> consistant to me.
>
> Thinking a bit more about this, its probably better to use this patch:
>
> diff --git a/arch/xtensa/lib/bootm.c b/arch/xtensa/lib/bootm.c
> index 6fcbfc4292..0e564507f9 100644
> --- a/arch/xtensa/lib/bootm.c
> +++ b/arch/xtensa/lib/bootm.c
> @@ -47,8 +47,8 @@ static struct bp_tag *setup_memory_tag(struct bp_tag 
> *params)
>         params->size = sizeof(struct meminfo);
>         mem = (struct meminfo *)params->data;
>         mem->type = MEMORY_TYPE_CONVENTIONAL;
> -       mem->start = gd->ram_base;
> -       mem->end = gd->ram_base + gd->ram_size;
> +       mem->start = PHYSADDR(gd->ram_base);
> +       mem->end = PHYSADDR(gd->ram_base + gd->ram_size);
>
>         printf("   MEMORY:          tag:0x%04x, type:0X%lx, 
> start:0X%lx, end:0X%lx\n",
>                BP_TAG_MEMORY, mem->type, mem->start, mem->end);
>
> What do you think?
>
Yes, I think this is better. arch_cpu_init runs earlier than 
setup_dest_addr in the boot process, so the gd->ram_base would have been 
overwritten anyway if set there.


Ovidiu

> Thanks,
> Stefan