[RESEND PATCH v3 4/5] mtd: nand: make 'erased check' optional
Boris Brezillon
boris.brezillon at free-electrons.com
Tue Sep 22 01:04:13 PDT 2015
Hi Brian,
On Mon, 21 Sep 2015 16:30:21 -0700
Brian Norris <computersforpeace at gmail.com> wrote:
> On Thu, Sep 03, 2015 at 06:03:41PM +0200, Boris Brezillon wrote:
> > Some ECC controllers do not need the extra 'erased check' done by the core.
> > Make it optional by creating a new NAND_ECC_DISABLE_ERASED_CHECK flag.
> >
> > Reed-Solomon ECC engines are guaranteed to generate ff ECC bytes when the
> > page in empty and are thus able to fix bitflips in erased pages.
> >
> > The software BCH implementation is also generating ff ECC bytes for empty
> > pages, and is thus able to fix bitflips in erased pages too.
> >
> > Signed-off-by: Boris Brezillon <boris.brezillon at free-electrons.com>
> > ---
> > drivers/mtd/nand/nand_base.c | 20 ++++++++++++++++----
> > drivers/mtd/nand/r852.c | 1 +
> > drivers/mtd/nand/tmio_nand.c | 1 +
> > drivers/mtd/nand/txx9ndfmc.c | 1 +
> > include/linux/mtd/nand.h | 8 ++++++++
> > 5 files changed, 27 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c
> > index 9a109a5..3be312d 100644
> > --- a/drivers/mtd/nand/nand_base.c
> > +++ b/drivers/mtd/nand/nand_base.c
> > @@ -1419,7 +1419,8 @@ static int nand_read_subpage(struct mtd_info *mtd, struct nand_chip *chip,
> >
> > stat = chip->ecc.correct(mtd, p,
> > &chip->buffers->ecccode[i], &chip->buffers->ecccalc[i]);
> > - if (stat == -EBADMSG) {
> > + if (stat == -EBADMSG &&
> > + !(chip->ecc.options & NAND_ECC_DISABLE_ERASED_CHECK)) {
> > /* check for empty pages with bitflips */
> > stat = nand_check_erased_ecc_chunk(p, chip->ecc.size,
> > &chip->buffers->ecccode[i],
> > @@ -1477,7 +1478,8 @@ static int nand_read_page_hwecc(struct mtd_info *mtd, struct nand_chip *chip,
> > int stat;
> >
> > stat = chip->ecc.correct(mtd, p, &ecc_code[i], &ecc_calc[i]);
> > - if (stat == -EBADMSG) {
> > + if (stat == -EBADMSG &&
> > + !(chip->ecc.options & NAND_ECC_DISABLE_ERASED_CHECK)) {
> > /* check for empty pages with bitflips */
> > stat = nand_check_erased_ecc_chunk(p, eccsize,
> > &ecc_code[i], eccbytes,
> > @@ -1537,7 +1539,8 @@ static int nand_read_page_hwecc_oob_first(struct mtd_info *mtd,
> > chip->ecc.calculate(mtd, p, &ecc_calc[i]);
> >
> > stat = chip->ecc.correct(mtd, p, &ecc_code[i], NULL);
> > - if (stat == -EBADMSG) {
> > + if (stat == -EBADMSG &&
> > + !(chip->ecc.options & NAND_ECC_DISABLE_ERASED_CHECK)) {
> > /* check for empty pages with bitflips */
> > stat = nand_check_erased_ecc_chunk(p, eccsize,
> > &ecc_code[i], eccbytes,
> > @@ -1600,7 +1603,8 @@ static int nand_read_page_syndrome(struct mtd_info *mtd, struct nand_chip *chip,
> > oob += chip->ecc.postpad;
> > }
> >
> > - if (stat == -EBADMSG) {
> > + if (stat == -EBADMSG &&
> > + !(chip->ecc.options & NAND_ECC_DISABLE_ERASED_CHECK)) {
> > /* check for empty pages with bitflips */
> > stat = nand_check_erased_ecc_chunk(p, chip->ecc.size,
> > oob - eccstepsize,
> > @@ -4300,6 +4304,14 @@ int nand_scan_tail(struct mtd_info *mtd)
> > ecc->write_oob_raw = ecc->write_oob;
> >
> > /*
> > + * The software implementation does not need the the erased check
> > + * since they already generate ff pattern for an erased page.
> > + */
> > + if (ecc->correct == nand_bch_correct_data ||
> > + ecc->correct == nand_correct_data)
> > + ecc->options |= NAND_ECC_DISABLE_ERASED_CHECK;
> > +
> > + /*
> > * The number of bytes available for a client to place data into
> > * the out of band area.
> > */
> > diff --git a/drivers/mtd/nand/r852.c b/drivers/mtd/nand/r852.c
> > index c9ad7a0..b82e4d9 100644
> > --- a/drivers/mtd/nand/r852.c
> > +++ b/drivers/mtd/nand/r852.c
> > @@ -876,6 +876,7 @@ static int r852_probe(struct pci_dev *pci_dev, const struct pci_device_id *id)
> > chip->ecc.hwctl = r852_ecc_hwctl;
> > chip->ecc.calculate = r852_ecc_calculate;
> > chip->ecc.correct = r852_ecc_correct;
> > + chip->ecc.options |= NAND_ECC_DISABLE_ERASED_CHECK;
>
> Maybe a comment here as to why? (e.g., something about hamming?)
Yep, I'll add a comment explaining why we put this flag.
[...]
> > diff --git a/include/linux/mtd/nand.h b/include/linux/mtd/nand.h
> > index 19d43b1..88956ab 100644
> > --- a/include/linux/mtd/nand.h
> > +++ b/include/linux/mtd/nand.h
> > @@ -128,6 +128,13 @@ typedef enum {
> > #define NAND_ECC_WRITE 1
> > /* Enable Hardware ECC before syndrome is read back from flash */
> > #define NAND_ECC_READSYN 2
> > +/*
> > + * Disable NAND 'page erased' check. In any case, this check is only done when
> > + * ecc.correct() returns -EBADMSG.
> > + * Set this flag if your implementation is able to fix bitflips in erased
> > + * pages.
> > + */
> > +#define NAND_ECC_DISABLE_ERASED_CHECK BIT(1)
>
> Any good reason you start at BIT(1) instead of BIT(0)?
Nope. I'll fix that too.
Best Regards,
Boris
--
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
More information about the linux-mtd
mailing list