Re: [PATCH v6 12/13] mtd: rawnand: brcmnand: Add support for getting ecc setting from strap

[Miquel Raynal]

Hi William,

william.zhang@xxxxxxxxxxxx wrote on Fri, 23 Feb 2024 09:25:09 -0800:

> Hi Miquel,
> 
> On 2/23/24 01:18, Miquel Raynal wrote:
> > Hi William,
> > 
> > william.zhang@xxxxxxxxxxxx wrote on Thu, 22 Feb 2024 19:47:57 -0800:
> >   
> >> BCMBCA broadband SoC based board design does not specify ecc setting in
> >> dts but rather use the SoC NAND strap info to obtain the ecc strength
> >> and spare area size setting. Add brcm,nand-ecc-use-strap dts propety for
> >> this purpose and update driver to support this option. However these two
> >> options can not be used at the same time.
> >>
> >> Signed-off-by: William Zhang <william.zhang@xxxxxxxxxxxx>
> >> Reviewed-by: David Regan <dregan@xxxxxxxxxxxx>
> >>  
> > 
> > FYI I did not receive patches 7, 8, 9, which makes the series numbering
> > very odd.
> >   
> I was using the get maintainer script mainly and it sends to the linux MTD list.  I will add your email directly next time.

Yes, I prefer to be in Cc of the whole series, please.

> >> ---
> >>
> >> Changes in v6:
> >> - Combine the ecc step size and ecc strength into one get function
> >> - Treat it as error condition if both brcm,nand-ecc-use-strap and nand
> >> ecc dts properties are set
> >> - Add intermediate steps to get the sector size bitfield
> >>
> >> Changes in v5: None
> >> Changes in v4:
> >> - Update the comments for ecc setting selection
> >>
> >> Changes in v3: None
> >> Changes in v2:
> >> - Minor cosmetic fixes
> >>
> >>   drivers/mtd/nand/raw/brcmnand/brcmnand.c | 83 ++++++++++++++++++++++--
> >>   1 file changed, 77 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/drivers/mtd/nand/raw/brcmnand/brcmnand.c b/drivers/mtd/nand/raw/brcmnand/brcmnand.c
> >> index ef7d340475be..e8ffc283b365 100644
> >> --- a/drivers/mtd/nand/raw/brcmnand/brcmnand.c
> >> +++ b/drivers/mtd/nand/raw/brcmnand/brcmnand.c
> >> @@ -1038,6 +1038,22 @@ static inline int brcmnand_sector_1k_shift(struct brcmnand_controller *ctrl)
> >>   		return -1;
> >>   }  
> >>   >> +static int brcmnand_get_sector_size_1k(struct brcmnand_host *host)  
> >> +{
> >> +	struct brcmnand_controller *ctrl = host->ctrl;
> >> +	int sector_size_bit = brcmnand_sector_1k_shift(ctrl);
> >> +	u16 acc_control_offs = brcmnand_cs_offset(ctrl, host->cs,
> >> +						  BRCMNAND_CS_ACC_CONTROL);
> >> +	u32 acc_control;
> >> +
> >> +	if (sector_size_bit < 0)
> >> +		return 0;
> >> +
> >> +	acc_control = nand_readreg(ctrl, acc_control_offs);
> >> +
> >> +	return (acc_control & BIT(sector_size_bit)) >> sector_size_bit;  
> > 
> > FIELD_PREP, FIELD_GET, *please*.  
> You probably missed my reply to your comments on the same patch in v5. Here is the link for the post in case it lost in your email:
> https://lore.kernel.org/lkml/c145b90c-e9f0-4d82-94cc-baf7bfda5954@xxxxxxxxx/T/#m1d911d2f119f3bd345c575a81b60bc2bd8c461eb

I didn't miss it, but the reason does not sound legitimate to me.
Please work on it, it will be so much cleaner.

> The mask is not constant here and cause build errors.
> >   
> >> +}
> >> +
> >>   static void brcmnand_set_sector_size_1k(struct brcmnand_host *host, int val)
> >>   {
> >>   	struct brcmnand_controller *ctrl = host->ctrl;
> >> @@ -1055,6 +1071,43 @@ static void brcmnand_set_sector_size_1k(struct brcmnand_host *host, int val)
> >>   	nand_writereg(ctrl, acc_control_offs, tmp);
> >>   }  
> >>   >> +static int brcmnand_get_spare_size(struct brcmnand_host *host)  
> >> +{
> >> +	struct brcmnand_controller *ctrl = host->ctrl;
> >> +	u16 acc_control_offs = brcmnand_cs_offset(ctrl, host->cs,
> >> +						  BRCMNAND_CS_ACC_CONTROL);
> >> +	u32 acc = nand_readreg(ctrl, acc_control_offs);
> >> +
> >> +	return (acc & brcmnand_spare_area_mask(ctrl));
> >> +}
> >> +
> >> +static void brcmnand_get_ecc_settings(struct brcmnand_host *host, struct nand_chip *chip)
> >> +{
> >> +	struct brcmnand_controller *ctrl = host->ctrl;
> >> +	u16 acc_control_offs = brcmnand_cs_offset(ctrl, host->cs,
> >> +						  BRCMNAND_CS_ACC_CONTROL);
> >> +	int sector_size_1k = brcmnand_get_sector_size_1k(host);
> >> +	int spare_area_size, ecc_level;
> >> +	u32 acc;
> >> +
> >> +	spare_area_size = brcmnand_get_spare_size(host);
> >> +	acc = nand_readreg(ctrl, acc_control_offs);
> >> +	ecc_level = (acc & brcmnand_ecc_level_mask(ctrl)) >> ctrl->ecc_level_shift;  
> > 
> > ditto
> >   
> >> +	if (sector_size_1k)
> >> +		chip->ecc.strength = ecc_level * 2;
> >> +	else if (spare_area_size == 16 && ecc_level == 15)
> >> +		chip->ecc.strength = 1; /* hamming */
> >> +	else
> >> +		chip->ecc.strength = ecc_level;
> >> +
> >> +	if (chip->ecc.size == 0) {
> >> +		if (sector_size_1k < 0)  
> > 
> > Should be <= 0 I guess
> >   
> >> +			chip->ecc.size = 512;
> >> +		else
> >> +			chip->ecc.size = 512 << sector_size_1k;  
> > 
> > What is this? Are you expecting sector_size_1k to be 0 or 1
> > and thus multiply 512 by two?
> >   
> Explained in the same post above. Sector_size_1k can be negative number for error condition where we default to 512 step size. Otherwise 0 for 512 and 1 for 1K which the above shift takes care of.

The logic is unclear, unnatural. Please simplify. You have the
possibility to change all the driver, so please simplify and clarify
the logic.

> > Please just use:
> > 			chip->ecc.size = SZ_1K;
> > 			  

Thanks,
Miquèl