[U-Boot] [PATCH v1 0/7] Enable high speed and heavy load for DDR4 for LSCH3

[York Sun]

On 11/05/2015 10:19 AM, Joakim Tjernlund wrote:
> On Thu, 2015-11-05 at 09:42 -0800, York Sun wrote:
>>
>> On 11/05/2015 01:55 AM, Joakim Tjernlund wrote:
>>> On Thu, 2015-11-05 at 08:23 +0000, Yuantian Tang wrote:
>>>> Hi Jocke,
>>>>
>>>> we achieved deep sleep mode that did exactly what you asked for.
>>>> If waken up from deep sleep, soc will resume from uboot and re-initialized DDR controller with contents
>>>> untouched.
>>>> Please refer to drivers/ddr/fsl/fsl_ddr_gen4.c and look at DEEP_SLEEP related code.
>>>
>>> Looking at it now and it looks the same as for ddr3? Some questions though:
>>>  289		if (is_warm_boot()) {
>>>  289                 /* enter self-refresh */
>>>  290                 temp_sdram_cfg = ddr_in32(&ddr->sdram_cfg_2);
>>>  291                 temp_sdram_cfg |= SDRAM_CFG2_FRC_SR;
>>>  292                 ddr_out32(&ddr->sdram_cfg_2, temp_sdram_cfg);
>>>
>>> Why do you need to force SR here? The DDR RAM must already be in SR at this point?
>>> I come from CPU reset state so my DDR controller has HW default values so
>>> this does not feel safe.
>>
>> This may be redundant. If the code runs to this line, it should come back from a
>> deep sleep. The core is in reset state but the DDR controller is not. It should
>> be in self-refresh mode. I will leave that to Yuantian to comment.
> 
> OK
> 
>>
>>>
>>>  293                 /* do board specific memory setup */
>>>  294                 board_mem_sleep_setup();
>>>  295 
>>>  296                 temp_sdram_cfg = (ddr_in32(&ddr->sdram_cfg) | SDRAM_CFG_BI);
>>> SDRAM_CFG_BI skips a lot(all?) init of DDR RAM. What if you want to change some DDR RAM
>>> timing/config due to a bug? Then you would have to force a cold start.
>>>
>>> Do you use ECC? Seems to be some issues with ECC if you skip D_INIT
>>>
>>
>> To perform a warm start, the data in DDR is preserved. So you don't need to init
>> the data again for ECC. To preserve data, you cannot run D_INIT again, which
>> will destroy the data for sure.
> 
> yes, but what about SDRAM_CFG_BI? Why do you need to set that here?
> I much rather just skip D_INIT and reconfigure DDR RAM, just in case one wants
> to correct some aspect of DDR config in later releases and feels a lot more robust.
> 
> Our systems cannot be coldstarted just because you upgrade SW on them.

Jocke,

If your memory has been intialized, you can set [BI] bit to bypass
re-initialization. It does different things than D_INIT. In short, D_INIT
initialize the data, i.e. the content of DDR, while BI bypassing initializing
DDR memory (or "chip" if that is easier to understand).

York