Messages in this thread | | | Subject | Re: [PATCH v7 2/2] mtd: rawnand: meson: add support for Amlogic NAND flash controller | From | Liang Yang <> | Date | Tue, 11 Dec 2018 17:39:51 +0800 |
| |
Hi Boris,
On 2018/12/11 16:39, Boris Brezillon wrote: > On Tue, 11 Dec 2018 09:56:25 +0800 > Liang Yang <liang.yang@amlogic.com> wrote: > >> Hi Miquel, >> >> On 2018/12/10 22:50, Miquel Raynal wrote: >>> Hi Liang, >>> >>> Liang Yang <liang.yang@amlogic.com> wrote on Mon, 10 Dec 2018 20:12:39 >>> +0800: >>> >>>> On 2018/12/10 19:38, Boris Brezillon wrote: >>>>> On Mon, 10 Dec 2018 19:23:46 +0800 >>>>> Liang Yang <liang.yang@amlogic.com> wrote: >>>>> >>>>>>>> + mtd->ecc_stats.failed++; >>>>>>>> + continue; >>>>>>>> + } >>>>>>>> + mtd->ecc_stats.corrected += ECC_ERR_CNT(*info); >>>>>>>> + bitflips = max_t(u32, bitflips, ECC_ERR_CNT(*info)); >>>>>>>> + } >>>>>>> >>>>>>> Are you sure you handle correctly empty pages with bf? >>>>>>> >> if scramble is enable, i would say yes here. >>>>>> when scramble is disabled, i am considering how to use the helper >>>>>> nand_check_erased_ecc_chunk, but it seems that i can't get the ecc >>>>>> bytes which is caculated by ecc engine.by the way, nfc dma doesn't send >>>>>> out the ecc parity bytes. >>>>> >>>>> Even if the ECC engine is disabled? >>>>> >>>> No. >>>> When ECC engine is disabled, it can read the ecc parity bytes ; but there is another problem that i need to consider how code struct looks better when reading error with ecc opened and then try to raw read. >>>> Is there a good idea? >>> >>> When reading with ECC enabled, in case of uncorrectable error you >>> must re-read without ECC, then check if the page is empty or not with >>> the core helpers (nand_check_erased_*()). >>> >>> Is this what you meant? >>> >> yes. when uncorrectable ECC error, i need firstly read out the ECC bytes >> without ECC engine and then use the helper nand_check_erased_ecc_chunk >> to check if blank page. >> Of course, the precondition is without scrambler, or the bland page can >> be detected by meson NFC. > > Yep, raw accessors should disable both the scrambler and the ECC > engine (see what's done in sunxi_nand.c). >
i see sunxi_nfc_hw_ecc_read_chunk and it will re-read the data for bitflips with scrambler off when ECC failed. also we can do the same implementation and it seems to be the only answer.
> . >
| |