Re: [PATCH v4 03/23] mtd: nand: add generic helpers to check, match, maximize ECC settings

From: Boris Brezillon
Date: Wed Jun 07 2017 - 02:16:20 EST


On Wed, 7 Jun 2017 10:48:33 +0900
Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote:

> 2017-06-07 6:47 GMT+09:00 Boris Brezillon <boris.brezillon@xxxxxxxxxxxxxxxxxx>:
> > On Tue, 6 Jun 2017 08:21:42 +0900
> > Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> >
> >> Driver are responsible for setting up ECC parameters correctly.
> >> Those include:
> >> - Check if ECC parameters specified (usually by DT) are valid
> >> - Meet the chip's ECC requirement
> >> - Maximize ECC strength if NAND_ECC_MAXIMIZE flag is set
> >>
> >> The logic can be generalized by factoring out common code.
> >>
> >> This commit adds 3 helpers to the NAND framework:
> >> nand_check_ecc_caps - Check if preset step_size and strength are valid
> >> nand_match_ecc_req - Match the chip's requirement
> >> nand_maximize_ecc - Maximize the ECC strength
> >>
> >> To use the helpers above, a driver needs to provide:
> >> - Data array of supported ECC step size and strength
> >> - A hook that calculates ECC bytes from the combination of
> >> step_size and strength.
> >>
> >> By using those helpers, code duplication among drivers will be
> >> reduced.
> >>
> >> Signed-off-by: Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx>
> >> ---
> >>
> >> Changes since the previous version:
> >>
> >> - Step size info holds an array of associated strengths
> >> - nand_match_ecc_req() does not take care of the case
> >> where ecc_size/strength is already set
> >> - Reflect more comments from Boris
> >>
> >> Previous version:
> >> http://patchwork.ozlabs.org/patch/752107/
> >>
> >>
> >> Changes in v4: None
> >> Changes in v3: None
> >> Changes in v2: None
> >>
> >> drivers/mtd/nand/nand_base.c | 219 +++++++++++++++++++++++++++++++++++++++++++
> >> include/linux/mtd/nand.h | 35 +++++++
> >> 2 files changed, 254 insertions(+)
> >>
> >> diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c
> >> index bdfa903..f2da4f2 100644
> >> --- a/drivers/mtd/nand/nand_base.c
> >> +++ b/drivers/mtd/nand/nand_base.c
> >> @@ -4509,6 +4509,225 @@ static int nand_set_ecc_soft_ops(struct mtd_info *mtd)
> >> }
> >> }
> >>
> >> +/**
> >> + * nand_check_ecc_caps - check the sanity of preset ECC settings
> >> + * @mtd: mtd info structure
> >> + * @chip: nand chip info structure
> >> + * @caps: ECC caps info structure
> >> + *
> >> + * When ECC step size and strength are already set, check if they are supported
> >> + * by the controller and the calculated ECC bytes fit within the chip's OOB.
> >> + * On success, the calculated ECC bytes is set.
> >> + */
> >> +int nand_check_ecc_caps(struct mtd_info *mtd, struct nand_chip *chip,

One more thing I didn't spot in my previous review: please only pass
chip here. mtd can be extracted using nand_to_mtd(chip). This is
applicable to all your helpers.

> >> + const struct nand_ecc_caps *caps)
> >> +{
> >> + const struct nand_ecc_step_info *stepinfo;
> >> + int avail_oobsize = mtd->oobsize - caps->oob_reserve_bytes;
> >> + int preset_step = chip->ecc.size;
> >> + int preset_strength = chip->ecc.strength;
> >> + int ecc_bytes;
> >> + int i, j;
> >> +
> >> + if (WARN_ON(avail_oobsize < 0))
> >> + return -EINVAL;
> >> +
> >> + if (!preset_step || !preset_strength)
> >> + return -ENODATA;
> >> +
> >> + for (i = 0; i < caps->nstepinfos; i++) {
> >> + stepinfo = &caps->stepinfos[i];
> >> +
> >> + if (stepinfo->stepsize != preset_step)
> >> + continue;
> >> +
> >> + for (j = 0; j < stepinfo->nstrengths; j++) {
> >> + if (stepinfo->strengths[j] == preset_strength)
> >> + goto found;
> >> + }
> >> + }
> >> +
> >> + pr_err("ECC (step, strength) = (%d, %d) not supported on this controller",
> >> + preset_step, preset_strength);
> >> +
> >> + return -ENOTSUPP;
> >> +
> >> +found:
> >
> > I prefer something like:
> >
> > if (i == caps->nstepinfos) {
> > pr_err(...);
> > return -ENOTSUPP;
> > }
> >
> > ...
> >
> > instead of this 'found' label.
>
>
> I want to bail-out if (step, strength) matches.
> In this version, the for-loop is double-nested by "step" and "strength".
> In C language, it is not possible to bail-out from multi-nested loop
> with a single "break;" statement. That is why I used "found:" label to do it.

You're right. I didn't pay attention to the nested for loop.

>
> In my first version where there was a single for-loop,
> I did not use the goto label.
> http://patchwork.ozlabs.org/patch/752107/
>
> Do you have any suggestion for cleaner implementation?
>
>

You can do:

nsteps = mtd->writesize / preset_step;

for (i = 0; i < caps->nstepinfos; i++) {
stepinfo = &caps->stepinfos[i];

if (stepinfo->stepsize != preset_step)
continue;

for (j = 0; j < stepinfo->nstrengths; j++) {
if (stepinfo->strengths[j] != preset_strength)
continue;

ecc_bytes = caps->calc_ecc_bytes(preset_step,
preset_strength);
if (WARN_ON_ONCE(ecc_bytes < 0))
return ecc_bytes;

if (ecc_bytes * nsteps > avail_oobsize) {
pr_err("ECC (step, strength) = (%d, %d) does not fit in OOB",
preset_step, preset_strength);
return -ENOSPC;
}

chip->ecc.bytes = ecc_bytes;

return 0;
}
}

pr_err("ECC (step, strength) = (%d, %d) not supported on this controller",
preset_step, preset_strength);

return -ENOTSUPP;


>
> >> + ecc_bytes = caps->calc_ecc_bytes(preset_step, preset_strength);
> >> + if (WARN_ON_ONCE(ecc_bytes < 0))
> >> + return ecc_bytes;
> >> +
> >> + if (ecc_bytes * mtd->writesize / preset_step > avail_oobsize) {
> >> + pr_err("ECC (step, strength) = (%d, %d) does not fit in OOB",
> >> + preset_step, preset_strength);
> >> + return -ENOSPC;
> >> + }
> >> +
> >> + chip->ecc.bytes = ecc_bytes;
> >> +
> >> + return 0;
> >> +}
> >> +EXPORT_SYMBOL_GPL(nand_check_ecc_caps);
> >> +
> >> +/**
> >> + * nand_match_ecc_req - meet the chip's requirement with least ECC bytes
> >> + * @mtd: mtd info structure
> >> + * @chip: nand chip info structure
> >> + * @caps: ECC engine caps info structure
> >> + *
> >> + * If a chip's ECC requirement is provided, try to meet it with the least
> >> + * number of ECC bytes (i.e. with the largest number of OOB-free bytes).
> >> + * On success, the chosen ECC settings are set.
> >> + */
> >> +int nand_match_ecc_req(struct mtd_info *mtd, struct nand_chip *chip,
> >> + const struct nand_ecc_caps *caps)
> >> +{
> >> + const struct nand_ecc_step_info *stepinfo;
> >> + int avail_oobsize = mtd->oobsize - caps->oob_reserve_bytes;
> >> + int req_step = chip->ecc_step_ds;
> >> + int req_strength = chip->ecc_strength_ds;
> >> + int req_corr, step_size, strength, steps, ecc_bytes, ecc_bytes_total;
> >> + int best_step, best_strength, best_ecc_bytes;
> >> + int best_ecc_bytes_total = INT_MAX;
> >
> > Just nitpicking, but why not -1 instead of INT_MAX?
>
> Because nand_match_ecc_req() prefers a smaller ecc_bytes_total.
> So I chose the largest int number as an init value.
> If we started from -1, the following if-conditional would have no effect.

Okay, that's a good reason :-).

>
> /*
> * We assume the best is to meet the chip's requrement
> * with the least number of ECC bytes.
> */
> if (ecc_bytes_total < best_ecc_bytes_total) {
> best_ecc_bytes_total = ecc_bytes_total;
> best_step = step_size;
> best_strength = strength;
> best_ecc_bytes = ecc_bytes;
> }
>
>
>
>
>
>
> >> + int i, j;
> >> +
> >> + if (WARN_ON(avail_oobsize < 0))
> >> + return -EINVAL;
> >> +
> >> + /* No information provided by the NAND chip */
> >> + if (!req_step || !req_strength)
> >> + return -ENOTSUPP;
> >> +
> >> + /* number of correctable bits the chip requires in a page */
> >> + req_corr = mtd->writesize / req_step * req_strength;
> >> +
> >> + for (i = 0; i < caps->nstepinfos; i++) {
> >> + stepinfo = &caps->stepinfos[i];
> >> + step_size = stepinfo->stepsize;
> >> +
> >> + for (j = 0; j < stepinfo->nstrengths; j++) {
> >> + strength = stepinfo->strengths[j];
> >> +
> >> + /*
> >> + * If both step size and strength are smaller than the
> >> + * chip's requirement, it is not easy to compare the
> >> + * resulted reliability.
> >> + */
> >> + if (step_size < req_step && strength < req_strength)
> >> + continue;
> >> +
> >> + if (mtd->writesize % step_size)
> >> + continue;
> >> +
> >> + steps = mtd->writesize / step_size;
> >> +
> >> + ecc_bytes = caps->calc_ecc_bytes(step_size, strength);
> >> + if (WARN_ON_ONCE(ecc_bytes < 0))
> >> + continue;
> >> + ecc_bytes_total = ecc_bytes * steps;
> >> +
> >> + if (ecc_bytes_total > avail_oobsize ||
> >> + strength * steps < req_corr)
> >> + continue;
> >> +
> >> + /*
> >> + * We assume the best is to meet the chip's requrement
> >> + * with the least number of ECC bytes.
> >> + */
> >> + if (ecc_bytes_total < best_ecc_bytes_total) {
> >> + best_ecc_bytes_total = ecc_bytes_total;
> >> + best_step = step_size;
> >> + best_strength = strength;
> >> + best_ecc_bytes = ecc_bytes;
> >> + }
> >> + }
> >> + }
> >> +
> >> + if (best_ecc_bytes_total == INT_MAX)
> >> + return -ENOTSUPP;
> >> +
> >> + chip->ecc.size = best_step;
> >> + chip->ecc.strength = best_strength;
> >> + chip->ecc.bytes = best_ecc_bytes;
> >> +
> >> + return 0;
> >> +}
> >> +EXPORT_SYMBOL_GPL(nand_match_ecc_req);
> >> +

[...]

> >> +
> >> +/**
> >> + * struct nand_ecc_caps - capability of ECC engine
> >> + * @stepinfos: array of ECC step information
> >> + * @nstepinfos: number of ECC step information
> >> + * @calc_ecc_bytes: driver's hook to calculate ECC bytes per step
> >> + * @oob_reserve_bytes: number of bytes in OOB that must be reserved
> >> + */
> >> +struct nand_ecc_caps {
> >> + const struct nand_ecc_step_info *stepinfos;
> >> + int nstepinfos;
> >> + int (*calc_ecc_bytes)(int step_size, int strength);
> >> + int oob_reserve_bytes;
> >
> > Why is this needed? I thought we agreed on passing oobavail as an
> > argument to these helper funcs. If a driver needs to reserve a few OOB
> > bytes, then doing mtd->oobsize - rsvd_bytes is not such a big deal.
>
>
> oobavail is really chip-dependent, so I agreed
> that it can not be included in the caps struct.
>
> Then, I flipped the logic.
> The number of reserved bytes will be more chip-independent.
> But, oob_reserve_bytes may not necessarily a fixed value.
>
> I can pass oobavail as a function argument.

Yes please.

Thanks,

Boris