mail archive of the barebox mailing list
 help / color / mirror / Atom feed
From: Ahmad Fatoum <a.fatoum@pengutronix.de>
To: David Jander <david@protonic.nl>
Cc: barebox@lists.infradead.org,
	Steffen Trumtrar <s.trumtrar@pengutronix.de>,
	Daniel Maslowski <info@orangecms.org>
Subject: Re: [PATCH 3/4] drivers: ddr: imx8m: ddr_init.c: support ECC scrubbing
Date: Wed, 4 Mar 2026 13:33:42 +0100	[thread overview]
Message-ID: <6377d7e5-ce81-4a8d-ab91-26da384ae5d9@pengutronix.de> (raw)
In-Reply-To: <20260304132322.0b0d956f@erd003.prtnl>

Hello David,

On 3/4/26 1:23 PM, David Jander wrote:
> On Wed, 4 Mar 2026 12:43:55 +0100
> Ahmad Fatoum <a.fatoum@pengutronix.de> wrote:
>>> +config IMX8MP_DRAM_ECC
>>> +	bool "Enable LPDDR4 ECC feature on i.MX8MP boards"
>>> +	depends on ARCH_IMX8MP
>>> +	help
>>> +	  The i.MX8MP SoC supports ECC on the LPDDR4 memory. Select Y to enable
>>> +	  this feature. The total amount of memory available will be reduced by
>>> +	  1/8th.  
>>
>> A generic option is too confusing when it's only affecting a single board.
> 
> Why should it affect only a single board? This is a general feature of this
> SoC and it can be enabled for any other board that uses this SoC with LPDDR4
> chips.

Sure and a user would expect that enabling this option would turn on
inline ECC, but it currently does so only for the Protonic board.
The i.MX8MP-EVK won't magically get ECC support with this enabled and I
think it may even fail to compile as it doesn't provide
board_dram_ecc_scrub on its own, so having this option in this form is
IMO confusing to users.

My suggestion below is thus to make ECC use a per-board (entry point)
decision instead of a build-wide decision.

> Even more so, I suspect this feature is present on some other SoCs as well
> which use the same Synopsys DesignWare IP with this feature enabled. In that
> case you'd probably want to make a more generic dram driver though. I suspect
> chips like the Rockchip RK3588 or RK3576 might support this also, but I
> haven't tried, and since those chips have binary dram init blobs this is
> probably not easy to figure out.

Interesting. While I have seen ECC documented for RK3568, I thought it
was not available for the RK3588.

By the way, Daniel (Cc'd) has been working on reverse engineering the
Rockchip DRAM init blob, so maybe that's moving within reach :)

Cheers,
Ahmad

> 
> Best regards,
> 
>> Alternative suggestion:
>>
>>   - make IMX8MP_DRAM_ECC a hidden symbol
>>   - select it from the protonic board
>>   - Turn the RAM timings of the protonic board into a header and
>>     replace #ifdef CONFIG_IMX8MP_DRAM_ECC with #ifdef USE_ECC
>>   - Add two source code files, that include the new header once with
>>     USE_ECC supported and once without
>>   - Add two entry points which only differ in what RAM timings are used
>>
>> Also worth considering: adding the size of the region to be scrubbed or
>> the function doing the scrubbing into dram_timing_info.
>>
>> Thoughts?
>>
>> Cheers,
>> Ahmad
>>
>>> +
>>>  endmenu
>>>  
>>>  endif
>>> diff --git a/drivers/ddr/imx/imx8m_ddr_init.c b/drivers/ddr/imx/imx8m_ddr_init.c
>>> index c16e04d274..6398a2b971 100644
>>> --- a/drivers/ddr/imx/imx8m_ddr_init.c
>>> +++ b/drivers/ddr/imx/imx8m_ddr_init.c
>>> @@ -45,6 +45,74 @@ static void ddr_cfg_umctl2(struct dram_controller *dram, struct dram_cfg_param *
>>>  	}
>>>  }
>>>  
>>> +#ifdef CONFIG_IMX8MP_DRAM_ECC
>>> +void ddrc_inline_ecc_scrub(unsigned int start_address,
>>> +			   unsigned int range_address)
>>> +{
>>> +	unsigned int tmp;
>>> +
>>> +	pr_debug("ECC scrub %08x-%08x\n", start_address, range_address);
>>> +	/* Step1: Enable quasi-dynamic programming */
>>> +	reg32_write(DDRC_SWCTL(0), 0x00000000);
>>> +	/* Step2: Set ECCCFG1.ecc_parity_region_lock to 1 */
>>> +	reg32setbit(DDRC_ECCCFG1(0), 0x4);
>>> +	/* Step3: Block the AXI ports from taking the transaction */
>>> +	reg32_write(DDRC_PCTRL_0(0), 0x0);
>>> +	/* Step4: Set scrub start address */
>>> +	reg32_write(DDRC_SBRSTART0(0), start_address);
>>> +	/* Step5: Set scrub range address */
>>> +	reg32_write(DDRC_SBRRANGE0(0), range_address);
>>> +	/* Step6: Set scrub_mode to write */
>>> +	reg32_write(DDRC_SBRCTL(0), 0x00000014);
>>> +	/* Step7: Set the desired pattern through SBRWDATA0 registers */
>>> +	reg32_write(DDRC_SBRWDATA0(0), 0x55aa55aa);
>>> +	/* Step8: Enable the SBR by programming SBRCTL.scrub_en=1 */
>>> +	reg32setbit(DDRC_SBRCTL(0), 0x0);
>>> +	/* Step9: Poll SBRSTAT.scrub_done=1 */
>>> +	tmp = reg32_read(DDRC_SBRSTAT(0));
>>> +	while (tmp != 0x00000002)
>>> +		tmp = reg32_read(DDRC_SBRSTAT(0)) & 0x2;
>>> +	/* Step10: Poll SBRSTAT.scrub_busy=0 */
>>> +	tmp = reg32_read(DDRC_SBRSTAT(0));
>>> +	while (tmp != 0x0)
>>> +		tmp = reg32_read(DDRC_SBRSTAT(0)) & 0x1;
>>> +	/* Step11: Disable SBR by programming SBRCTL.scrub_en=0 */
>>> +	clrbits_le32(DDRC_SBRCTL(0), 0x1);
>>> +	/* Step12: Prepare for normal scrub operation(Read) and set scrub_interval*/
>>> +	reg32_write(DDRC_SBRCTL(0), 0xff20);
>>> +	/* Step13: Enable the SBR by programming SBRCTL.scrub_en=1 */
>>> +	reg32_write(DDRC_SBRCTL(0), 0xff21);
>>> +	/* Step14: Enable AXI ports by programming */
>>> +	reg32_write(DDRC_PCTRL_0(0), 0x1);
>>> +	/* Step15: Disable quasi-dynamic programming */
>>> +	reg32_write(DDRC_SWCTL(0), 0x00000001);
>>> +}
>>> +
>>> +void ddrc_inline_ecc_scrub_end(unsigned int start_address,
>>> +			       unsigned int range_address)
>>> +{
>>> +	pr_debug("ECC  end %08x-%08x\n", start_address, range_address);
>>> +	/* Step1: Enable quasi-dynamic programming */
>>> +	reg32_write(DDRC_SWCTL(0), 0x00000000);
>>> +	/* Step2: Block the AXI ports from taking the transaction */
>>> +	reg32_write(DDRC_PCTRL_0(0), 0x0);
>>> +	/* Step3: Set scrub start address */
>>> +	reg32_write(DDRC_SBRSTART0(0), start_address);
>>> +	/* Step4: Set scrub range address */
>>> +	reg32_write(DDRC_SBRRANGE0(0), range_address);
>>> +	/* Step5: Disable SBR by programming SBRCTL.scrub_en=0 */
>>> +	clrbits_le32(DDRC_SBRCTL(0), 0x1);
>>> +	/* Step6: Prepare for normal scrub operation(Read) and set scrub_interval */
>>> +	reg32_write(DDRC_SBRCTL(0), 0x100);
>>> +	/* Step7: Enable the SBR by programming SBRCTL.scrub_en=1 */
>>> +	reg32_write(DDRC_SBRCTL(0), 0x101);
>>> +	/* Step8: Enable AXI ports by programming */
>>> +	reg32_write(DDRC_PCTRL_0(0), 0x1);
>>> +	/* Step9: Disable quasi-dynamic programming */
>>> +	reg32_write(DDRC_SWCTL(0), 0x00000001);
>>> +}
>>> +#endif
>>> +
>>>  static unsigned int g_cdd_rr_max[4];
>>>  static unsigned int g_cdd_rw_max[4];
>>>  static unsigned int g_cdd_wr_max[4];
>>> @@ -642,6 +710,11 @@ int imx8m_ddr_init(struct dram_controller *dram, struct dram_timing_info *dram_t
>>>  	reg32_write(DDRC_PCTRL_0(0), 0x00000001);
>>>  	pr_debug("ddrmix config done\n");
>>>  
>>> +#ifdef CONFIG_IMX8MP_DRAM_ECC
>>> +	if (dram->ddrc_type == DDRC_TYPE_MP)
>>> +		board_dram_ecc_scrub();
>>> +#endif
>>> +
>>>  	/* save the dram timing config into memory */
>>>  	dram_config_save(dram, dram_timing, IMX8M_SAVED_DRAM_TIMING_BASE);
>>>  
>>> diff --git a/include/soc/imx8m/ddr.h b/include/soc/imx8m/ddr.h
>>> index 5df07772b3..08ebf61da0 100644
>>> --- a/include/soc/imx8m/ddr.h
>>> +++ b/include/soc/imx8m/ddr.h
>>> @@ -186,6 +186,8 @@
>>>  #define DDRC_SBRWDATA0(X)        (DDRC_IPS_BASE_ADDR(X) + 0xf2c)
>>>  #define DDRC_SBRWDATA1(X)        (DDRC_IPS_BASE_ADDR(X) + 0xf30)
>>>  #define DDRC_PDCH(X)             (DDRC_IPS_BASE_ADDR(X) + 0xf34)
>>> +#define DDRC_SBRSTART0(X)        (DDRC_IPS_BASE_ADDR(X) + 0xf38)
>>> +#define DDRC_SBRRANGE0(X)        (DDRC_IPS_BASE_ADDR(X) + 0xf40)
>>>  
>>>  #define DDRC_FREQ1_DERATEEN(X)         (DDRC_IPS_BASE_ADDR(X) + 0x2020)
>>>  #define DDRC_FREQ1_DERATEINT(X)        (DDRC_IPS_BASE_ADDR(X) + 0x2024)
>>> @@ -380,4 +382,13 @@ static inline void imx8m_ddr_load_train_code(enum dram_type dram_type,
>>>  	ddr_load_train_code(&imx8m_dram_controller, dram_type, fw_type);
>>>  }
>>>  
>>> +#define DDRC_PHY_REG(x)	((x) * 4)
>>> +
>>> +void board_dram_ecc_scrub(void);
>>> +
>>> +void ddrc_inline_ecc_scrub(unsigned int start_address,
>>> +			   unsigned int range_address);
>>> +void ddrc_inline_ecc_scrub_end(unsigned int start_address,
>>> +			       unsigned int range_address);
>>> +
>>>  #endif
>>>   
>>
> 
> 
> 

-- 
Pengutronix e.K.                  |                             |
Steuerwalder Str. 21              | http://www.pengutronix.de/  |
31137 Hildesheim, Germany         | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686  | Fax:   +49-5121-206917-5555 |




  reply	other threads:[~2026-03-04 12:34 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-03-04 11:23 [PATCH 0/4] ARM: i.MX8: add DDRC-ECC support Steffen Trumtrar
2026-03-04 11:23 ` [PATCH 1/4] ARM: i.MX: esdctl: fix spelling of ad(d)ress Steffen Trumtrar
2026-03-04 11:23 ` [PATCH 2/4] arm: mach-imx: esdctl.c: Add support for imx8mp inline ECC Steffen Trumtrar
2026-03-04 11:23 ` [PATCH 3/4] drivers: ddr: imx8m: ddr_init.c: support ECC scrubbing Steffen Trumtrar
2026-03-04 11:43   ` Ahmad Fatoum
2026-03-04 12:23     ` David Jander
2026-03-04 12:33       ` Ahmad Fatoum [this message]
2026-03-04 13:14         ` David Jander
2026-03-04 11:23 ` [PATCH 4/4] arm: boards: protonic-imx8ml: Add ECC + scrubbing Steffen Trumtrar
2026-03-04 11:34   ` Ahmad Fatoum

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=6377d7e5-ce81-4a8d-ab91-26da384ae5d9@pengutronix.de \
    --to=a.fatoum@pengutronix.de \
    --cc=barebox@lists.infradead.org \
    --cc=david@protonic.nl \
    --cc=info@orangecms.org \
    --cc=s.trumtrar@pengutronix.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox