[PATCH 01/11] imx: implement get_effective_memsize

[Peng Fan]

On 11/9/2022 3:40 PM, Pali Rohár wrote:
> On Wednesday 09 November 2022 09:48:53 Peng Fan wrote:
>> On 11/8/2022 4:03 PM, Pali Rohár wrote:
>>> On Tuesday 08 November 2022 07:56:59 Peng Fan wrote:
>>>>> Subject: Re: [PATCH 01/11] imx: implement get_effective_memsize
>>>>>
>>>>> On Tuesday 08 November 2022 09:38:01 Peng Fan wrote:
>>>>>> On 11/7/2022 3:55 PM, Pali Rohár wrote:
>>>>>>> On Monday 07 November 2022 16:00:06 Peng Fan (OSS) wrote:
>>>>>>>> From: Peng Fan <peng.fan at nxp.com>
>>>>>>>>
>>>>>>>> To i.MX6/7 which has 2GB memory, the upper 4KB cut off, will cause
>>>>>>>> the top 1MB not mapped as normal memory, because ARMV7-A use
>>>>>>>> section mapping. So implement i.MX6/7 specific
>>>>>>>> get_effective_memsize to fix the issue.
>>>>>>>>
>>>>>>>> Fixes: 777aaaa706bc("common/memsize.c: Fix get_effective_memsize()
>>>>>>>> to check for overflow")
>>>>>>>> Signed-off-by: Peng Fan <peng.fan at nxp.com>
>>>>>>>
>>>>>>> Should not just configuring CONFIG_MAX_MEM_MAPPED properly avoid
>>>>> that issue?
>>>>>>
>>>>>> No, unless I decrease PHYS_SDRAM_SIZE.
>>>>>
>>>>> So, what is the issue? I just do not see what happens after 777aaaa706bc
>>>>> that RAM size is calculated incorrectly in your case. I did not catch the
>>>>> description from commit message. What are your gd->ram_size and
>>>>> CONFIG_MAX_MEM_MAPPED values that current code does not work?
>>>>
>>>> The base is 2GB, the size is 2GB. With CONFIG_MAX_MEM_MAPPED,
>>>> the ram_size already decreased by 4KB.
>>>
>>> If base is 2GB and size is 2GB then ram_top is 4GB which cannot be
>>> represented in 32-bit phys_size_t type and hence some of other u-boot
>>> functions use 0 as ram_top. Mentioned commit tries to fix this issue.
>>> I guess that you have some other hidden problem and my change just
>>> showed implication of that.
>>
>> The issue is with higher 4KB cut off, the MMU mapping will cut off
>> 1MB or 2MB(section mapping), so after MMU enabled, the PC instruction will
>> not able to fetch instruction in the higher 1MB area because of U-Boot
>> relocated to
>> top of DRAM.
> 
> But it is not possible to represent whole 4GB of RAM size in 32-bit
> phys_size_t type. So some cut-off for this storage is required.

I understand this. But this means I have to reserve the higher 2MB 
section because of the 4KB cut off.

Even the ram_top is 0 in my case, it does not matter. The relocaddr
is unsigned long, when reserve other memory, it will back to the
high address.

Regards,
Peng.

> 
>> Regards,
>> Peng.
>>
>> Could you check how is gd->ram_top
>>> configured with and without this change?
> 
> Could you check this?
> 
>>>> Regards,
>>>> Peng.
>>>>
>>>>>
>>>>>> Regards,
>>>>>> Peng.
>>>>>>
>>>>>>>
>>>>>>>> ---
>>>>>>>>     arch/arm/mach-imx/cache.c | 14 ++++++++++++++
>>>>>>>>     1 file changed, 14 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/arch/arm/mach-imx/cache.c b/arch/arm/mach-imx/cache.c
>>>>>>>> index ab9b621a2a6..69a085abee7 100644
>>>>>>>> --- a/arch/arm/mach-imx/cache.c
>>>>>>>> +++ b/arch/arm/mach-imx/cache.c
>>>>>>>> @@ -7,10 +7,24 @@
>>>>>>>>     #include <cpu_func.h>
>>>>>>>>     #include <asm/armv7.h>
>>>>>>>>     #include <asm/cache.h>
>>>>>>>> +#include <asm/global_data.h>
>>>>>>>>     #include <asm/pl310.h>
>>>>>>>>     #include <asm/io.h>
>>>>>>>>     #include <asm/mach-imx/sys_proto.h>
>>>>>>>> +DECLARE_GLOBAL_DATA_PTR;
>>>>>>>> +
>>>>>>>> +phys_size_t get_effective_memsize(void) { #ifndef
>>>>>>>> +CONFIG_MAX_MEM_MAPPED
>>>>>>>> +	return gd->ram_size;
>>>>>>>> +#else
>>>>>>>> +	/* limit stack to what we can reasonable map */
>>>>>>>> +	return ((gd->ram_size > CONFIG_MAX_MEM_MAPPED) ?
>>>>>>>> +		CONFIG_MAX_MEM_MAPPED : gd->ram_size); #endif }
>>>>>>>> +
>>>>>>>>     void enable_ca7_smp(void)
>>>>>>>>     {
>>>>>>>>     	u32 val;
>>>>>>>> --
>>>>>>>> 2.36.0
>>>>>>>>