Re: [PATCH 8/9] mtd: fsl-quadspi: disable AHB buffer prefetch

From: Han Xu
Date: Fri Apr 22 2016 - 00:49:15 EST




________________________________________
From: Yunhui Cui
Sent: Thursday, April 21, 2016 9:52 PM
To: Han Xu; Yunhui Cui; dwmw2@xxxxxxxxxxxxx; computersforpeace@xxxxxxxxx; han.xu@xxxxxxxxxxxxx
Cc: linux-kernel@xxxxxxxxxxxxxxx; linux-mtd@xxxxxxxxxxxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; Yao Yuan
Subject: RE: [PATCH 8/9] mtd: fsl-quadspi: disable AHB buffer prefetch

> -----Original Message-----
> From: Han Xu
> Sent: Thursday, April 21, 2016 11:48 PM
> To: Yunhui Cui; Yunhui Cui; dwmw2@xxxxxxxxxxxxx;
> computersforpeace@xxxxxxxxx; han.xu@xxxxxxxxxxxxx
> Cc: linux-kernel@xxxxxxxxxxxxxxx; linux-mtd@xxxxxxxxxxxxxxxxxxx; linux-
> arm-kernel@xxxxxxxxxxxxxxxxxxx; Yao Yuan
> Subject: Re: [PATCH 8/9] mtd: fsl-quadspi: disable AHB buffer prefetch
>
>
>
> ________________________________________
> From: Yunhui Cui
> Sent: Thursday, April 21, 2016 3:41 AM
> To: Han Xu; Yunhui Cui; dwmw2@xxxxxxxxxxxxx; computersforpeace@xxxxxxxxx;
> han.xu@xxxxxxxxxxxxx
> Cc: linux-kernel@xxxxxxxxxxxxxxx; linux-mtd@xxxxxxxxxxxxxxxxxxx; linux-
> arm-kernel@xxxxxxxxxxxxxxxxxxx; Yao Yuan
> Subject: RE: [PATCH 8/9] mtd: fsl-quadspi: disable AHB buffer prefetch
>
> On Thu, Apr 24, 2016 at 06:37:01 AM +0800, Han Xu wrote:
> > ________________________________________
> > From: Yunhui Cui <B56489@xxxxxxxxxxxxx>
> > Sent: Wednesday, April 13, 2016 9:50 PM
> > To: dwmw2@xxxxxxxxxxxxx; computersforpeace@xxxxxxxxx;
> > han.xu@xxxxxxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx; linux-mtd@xxxxxxxxxxxxxxxxxxx;
> > linux- arm-kernel@xxxxxxxxxxxxxxxxxxx; Yao Yuan; Yunhui Cui
> > Subject: [PATCH 8/9] mtd: fsl-quadspi: disable AHB buffer prefetch
> >
> > From: Yunhui Cui <yunhui.cui@xxxxxxx>
> >
> > A-009282: QuadSPI: QuadSPI data pre-fetch can result in incorrect data
> > Affects: QuadSPI
> > Description: With AHB buffer prefetch enabled, the QuadSPI may return
> > incorrect data on the AHB interface. The buffer pre-fetch is enabled
> > if the fetch size as configured either in the LUT or in the BUFxCR
> > register is greater than 8 bytes.
> > Impact: Only 64 bit read allowed.
> > Workaround: Keep the read data size to 64 bits (8 Bytes), which
> > disables the prefetch on the AHB buffer, and prevents this issue from
> occurring.
> >
> > Signed-off-by: Yunhui Cui <yunhui.cui@xxxxxxx>
> > ---
> > drivers/mtd/spi-nor/fsl-quadspi.c | 29 +++++++++++++++++++++++------
> > 1 file changed, 23 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/mtd/spi-nor/fsl-quadspi.c
> > b/drivers/mtd/spi-nor/fsl- quadspi.c index fea18b6..d9f3e50 100644
> > --- a/drivers/mtd/spi-nor/fsl-quadspi.c
> > +++ b/drivers/mtd/spi-nor/fsl-quadspi.c
> > @@ -752,19 +752,36 @@ static void fsl_qspi_init_abh_read(struct
> > fsl_qspi
> > *q) {
> > void __iomem *base = q->iobase;
> > int seqid;
> > + const struct fsl_qspi_devtype_data *devtype_data =
> > + q->devtype_data;
> >
> > /* AHB configuration for access buffer 0/1/2 .*/
> > qspi_writel(q, QUADSPI_BUFXCR_INVALID_MSTRID, base +
> > QUADSPI_BUF0CR);
> > qspi_writel(q, QUADSPI_BUFXCR_INVALID_MSTRID, base +
> > QUADSPI_BUF1CR);
> > qspi_writel(q, QUADSPI_BUFXCR_INVALID_MSTRID, base +
> > QUADSPI_BUF2CR);
> > +
> > /*
> > - * Set ADATSZ with the maximum AHB buffer size to improve the
> > - * read performance.
> > + * Errata: A-009282: QuadSPI data prefetch may result in
> > incorrect data
> > + * Workaround: Keep the read data size to 64 bits (8 bytes).
> > + * This disables the prefetch on the AHB buffer and
> > + * prevents this issue from occurring.
> > */
> > - qspi_writel(q, QUADSPI_BUF3CR_ALLMST_MASK |
> > - ((q->devtype_data->ahb_buf_size / 8)
> > - << QUADSPI_BUF3CR_ADATSZ_SHIFT),
> > - base + QUADSPI_BUF3CR);
> > + if (devtype_data->devtype == FSL_QUADSPI_LS2080A ||
> > + devtype_data->devtype == FSL_QUADSPI_LS1021A) {
> >
> > Use QUIRK, not SoC name.
> [Yunhui] This is a SoC errata, we need distinguish according to the
> devtype_data->devtype
>
> It's the same thing and we have already done that before.

> [Yunhui] your mean that I should add such as the following code:
#define QUADSPI_QUIRK_ERRATA_9282 (1 << 4)
static inline int has_errata_num_a009282(struct fsl_qspi *q)
{
return q->devtype_data->driver_data & QUADSPI_QUIRK_ERRATA_9282;
}

I mean QUIRK can be used even for HW limitation, but it's not necessary to
use ERRATA number as QUIRK name, you can name it with natural language,
such as QUADSPI_QUIRK_ADASZ_8BYTE_LIMIT, and describe the reason in
comments.

> /*
> * TKT253890, Controller needs driver to fill txfifo till 16 byte to
> * trigger data transfer even though extra data will not transferred.
> */
> #define QUADSPI_QUIRK_TKT253890 (1 << 2)
> /* Controller cannot wake up from wait mode, TKT245618 */
> #define QUADSPI_QUIRK_TKT245618 (1 << 3)
>
>
> > +
> > + qspi_writel(q, QUADSPI_BUF3CR_ALLMST_MASK |
> > + (1 << QUADSPI_BUF3CR_ADATSZ_SHIFT),
> > + base + QUADSPI_BUF3CR);
> > +
> > + } else {
> > + /*
> > + * Set ADATSZ with the maximum AHB buffer size to
> > + improve
> > the
> > + * read performance.
> > + */
> > + qspi_writel(q, QUADSPI_BUF3CR_ALLMST_MASK |
> > + ((q->devtype_data->ahb_buf_size / 8)
> > + << QUADSPI_BUF3CR_ADATSZ_SHIFT),
> > + base + QUADSPI_BUF3CR);
> > + }
> >
> > /* We only use the buffer3 */
> > qspi_writel(q, 0, base + QUADSPI_BUF0IND);
> > --
> > 2.1.0.27.g96db324