[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1ba00a38-7293-4f72-9aee-f87f41a3dcc6@gmail.com>
Date: Mon, 4 Aug 2025 09:40:54 +0200
From: Gabor Juhos <j4g8y7@...il.com>
To: Konrad Dybcio <konrad.dybcio@....qualcomm.com>,
Mark Brown <broonie@...nel.org>, Md Sadre Alam <quic_mdalam@...cinc.com>,
Varadarajan Narayanan <quic_varada@...cinc.com>,
Sricharan Ramabadhran <quic_srichara@...cinc.com>
Cc: linux-spi@...r.kernel.org, linux-mtd@...ts.infradead.org,
linux-arm-msm@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] spi: spi-qpic-snand: use correct CW_PER_PAGE value for
OOB write
Hi Konrad,
2025. 08. 01. 13:08 keltezéssel, Konrad Dybcio írta:
...
>> diff --git a/drivers/spi/spi-qpic-snand.c b/drivers/spi/spi-qpic-snand.c
>> index 0cfa0d960fd3c245c2bbf4f5e02d0fc0b13e7696..5216d60e01aab26f927baaea24296571a77527cb 100644
>> --- a/drivers/spi/spi-qpic-snand.c
>> +++ b/drivers/spi/spi-qpic-snand.c
>> @@ -1196,7 +1196,7 @@ static int qcom_spi_program_oob(struct qcom_nand_controller *snandc,
>> u32 cfg0, cfg1, ecc_bch_cfg, ecc_buf_cfg;
>>
>> cfg0 = (ecc_cfg->cfg0 & ~CW_PER_PAGE_MASK) |
>> - FIELD_PREP(CW_PER_PAGE_MASK, num_cw - 1);
>> + FIELD_PREP(CW_PER_PAGE_MASK, 0);
>
> FWIW I'm just a fly-by reviewer for this driver, but the docs say:
>
> The value is the number of codewords per page minus one
Well, the driver uses that differently even without the patch. See below.
> "NOTE: This field must be cleared for block erase operation"
$ git grep -hp 'FIELD_PREP(CW_PER_PAGE_MASK,.*;' drivers/spi/spi-qpic-snand.c
static int qcom_spi_block_erase(struct qcom_nand_controller *snandc)
FIELD_PREP(CW_PER_PAGE_MASK, 0));
This function implements the block erase operation and it corresponds to the
documentation. So far so good.
static int qcom_spi_read_last_cw(struct qcom_nand_controller *snandc,
FIELD_PREP(CW_PER_PAGE_MASK, 0);
static int qcom_spi_read_cw_raw(struct qcom_nand_controller *snandc, u8 *data_buf,
FIELD_PREP(CW_PER_PAGE_MASK, 0);
These two functions are using a single codeword (with zero CW_PER_PAGE value).
So, it seems that in reality the CW_PER_PAGE value means the number of codewords
(minus one) used within a single operation executed. Of course it is possible
that the existing code is wrong here.
static int qcom_spi_read_page_ecc(struct qcom_nand_controller *snandc,
FIELD_PREP(CW_PER_PAGE_MASK, num_cw - 1);
static int qcom_spi_read_page_oob(struct qcom_nand_controller *snandc,
FIELD_PREP(CW_PER_PAGE_MASK, num_cw - 1);
static int qcom_spi_program_raw(struct qcom_nand_controller *snandc,
FIELD_PREP(CW_PER_PAGE_MASK, num_cw - 1);
static int qcom_spi_program_ecc(struct qcom_nand_controller *snandc,
FIELD_PREP(CW_PER_PAGE_MASK, num_cw - 1);
The previous functions are operating on whole pages, so those are using all codewords
within a page thus 'num_cw - 1' is getting set in the register field. This also matches
with the documentation.
static int qcom_spi_program_oob(struct qcom_nand_controller *snandc,
FIELD_PREP(CW_PER_PAGE_MASK, num_cw - 1);
This is the function fixed by the patch. As it is indicated in the commit description
this also uses a single codeword similarly to the qcom_spi_read_(last_cw,cw_raw) functions
described above so the CW_PER_PAGE value should be set to zero.
Regards,
Gabor
Powered by blists - more mailing lists