[PATCH 2/3] mtd: rawnand: rzn1: Add new NAND controller driver
Miquel Raynal
miquel.raynal at bootlin.com
Fri Nov 19 01:23:40 PST 2021
Hi Geert,
geert at linux-m68k.org wrote on Fri, 19 Nov 2021 09:55:53 +0100:
> Hi Miquel,
>
> CC Gareth
>
> On Thu, Nov 18, 2021 at 12:19 PM Miquel Raynal
> <miquel.raynal at bootlin.com> wrote:
> > Introduce Renesas RZ/N1x NAND controller driver which supports:
> > - All ONFI timing modes
> > - Different configurations of its internal ECC controller
> > - On-die (not tested) and software ECC support
> > - Several chips (not tested)
> > - Subpage accesses
> > - DMA and PIO
> >
> > Signed-off-by: Miquel Raynal <miquel.raynal at bootlin.com>
>
> Thanks for your patch!
>
> > --- a/drivers/mtd/nand/raw/Kconfig
> > +++ b/drivers/mtd/nand/raw/Kconfig
> > @@ -467,6 +467,12 @@ config MTD_NAND_PL35X
> > Enables support for PrimeCell SMC PL351 and PL353 NAND
> > controller found on Zynq7000.
> >
> > +config MTD_NAND_RZN1
> > + tristate "Renesas RZ/N1D, RZ/N1S, RZ/N1L NAND controller"
> > + depends on OF || COMPILE_TEST
>
> depends on ARCH_RENESAS || COMPILE_TEST
Yeah of course, sorry about that.
>
> > + help
> > + Enables support for Renesas RZ/N1x SoC family NAND controller.
> > +
> > comment "Misc"
> >
> > config MTD_SM_COMMON
>
> > --- /dev/null
> > +++ b/drivers/mtd/nand/raw/rzn1-nand-controller.c
>
> > +static int rzn1_read_subpage_hw_ecc(struct nand_chip *chip, u32 req_offset,
> > + u32 req_len, u8 *bufpoi, int page)
> > +{
> > + struct rzn1_nfc *nfc = to_rzn1_nfc(chip->controller);
> > + struct mtd_info *mtd = nand_to_mtd(chip);
> > + struct rzn1_nand_chip *rzn1_nand = to_rzn1_nand(chip);
> > + unsigned int cs = to_nfc_cs(rzn1_nand);
> > + unsigned int page_off = round_down(req_offset, chip->ecc.size);
> > + unsigned int real_len = round_up(req_offset + req_len - page_off,
> > + chip->ecc.size);
> > + unsigned int start_chunk = page_off / chip->ecc.size;
> > + unsigned int nchunks = real_len / chip->ecc.size;
> > + unsigned int ecc_off = 2 + (start_chunk * chip->ecc.bytes);
> > + struct rzn1_op rop = {
> > + .command = COMMAND_INPUT_SEL_AHBS | COMMAND_0(NAND_CMD_READ0) |
> > + COMMAND_2(NAND_CMD_READSTART) | COMMAND_FIFO_SEL |
> > + COMMAND_SEQ_READ_PAGE,
> > + .addr0_row = page,
> > + .addr0_col = page_off,
> > + .len = real_len,
> > + .ecc_offset = ECC_OFFSET(mtd->writesize + ecc_off),
> > + };
> > + unsigned int max_bitflips = 0;
> > + u32 ecc_stat;
> > + int bf, ret, i;
>
> unsigned int i
Strangely I'm used to always set my loop indexes as signed integers,
but I'll happily change that everywhere in the driver before
re-submitting.
[...]
> > +static int rzn1_write_page_hw_ecc(struct nand_chip *chip, const u8 *buf,
> > + int oob_required, int page)
> > +{
> > + struct rzn1_nfc *nfc = to_rzn1_nfc(chip->controller);
> > + struct mtd_info *mtd = nand_to_mtd(chip);
> > + struct rzn1_nand_chip *rzn1_nand = to_rzn1_nand(chip);
> > + unsigned int cs = to_nfc_cs(rzn1_nand);
> > + struct rzn1_op rop = {
> > + .command = COMMAND_INPUT_SEL_DMA | COMMAND_0(NAND_CMD_SEQIN) |
> > + COMMAND_1(NAND_CMD_PAGEPROG) | COMMAND_FIFO_SEL |
> > + COMMAND_SEQ_WRITE_PAGE,
> > + .addr0_row = page,
> > + .len = mtd->writesize,
> > + .ecc_offset = ECC_OFFSET(mtd->writesize + 2),
> > + };
> > + dma_addr_t dma_addr;
> > + int ret;
> > +
> > + memcpy(nfc->buf, buf, mtd->writesize);
> > +
> > + /* Prepare controller */
> > + rzn1_nfc_select_target(chip, chip->cur_cs);
> > + rzn1_nfc_clear_status(nfc);
> > + reinit_completion(&nfc->complete);
> > + rzn1_nfc_en_interrupts(nfc, INT_MEM_RDY(cs));
> > + rzn1_nfc_en_correction(nfc);
> > +
> > + /* Configure DMA */
> > + dma_addr = dma_map_single(nfc->dev, (void *)nfc->buf, mtd->writesize,
> > + DMA_TO_DEVICE);
> > + writel(dma_addr, nfc->regs + DMA_ADDR_LOW_REG);
> > + writel(mtd->writesize, nfc->regs + DMA_CNT_REG);
> > + writel(DMA_TLVL_MAX, nfc->regs + DMA_TLVL_REG);
> > +
> > + rzn1_nfc_trigger_op(nfc, &rop);
> > + rzn1_nfc_trigger_dma(nfc);
> > +
> > + ret = rzn1_nfc_wait_end_of_io(nfc, chip);
> > + dma_unmap_single(nfc->dev, dma_addr, mtd->writesize, DMA_TO_DEVICE);
> > + rzn1_nfc_dis_correction(nfc);
> > + if (ret) {
> > + dev_err(nfc->dev, "Write page operation never ending\n");
> > + return ret;
> > + }
> > +
> > + if (oob_required) {
>
> Return early if !oob_required, to reduce indentation below?
Yeah sure.
> > + ret = nand_change_write_column_op(chip, mtd->writesize,
> > + chip->oob_poi, mtd->oobsize,
> > + false);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > + return 0;
> > +}
[...]
> > +static int rzn1_nfc_probe(struct platform_device *pdev)
> > +{
> > + struct rzn1_nfc *nfc;
> > + int irq, ret;
> > +
> > + nfc = devm_kzalloc(&pdev->dev, sizeof(*nfc), GFP_KERNEL);
> > + if (!nfc)
> > + return -ENOMEM;
> > +
> > + nfc->dev = &pdev->dev;
> > + nand_controller_init(&nfc->controller);
> > + nfc->controller.ops = &rzn1_nfc_ops;
> > + INIT_LIST_HEAD(&nfc->chips);
> > + init_completion(&nfc->complete);
> > +
> > + nfc->regs = devm_platform_ioremap_resource(pdev, 0);
> > + if (IS_ERR(nfc->regs))
> > + return PTR_ERR(nfc->regs);
> > +
> > + rzn1_nfc_dis_interrupts(nfc);
> > + irq = platform_get_irq(pdev, 0);
>
> platform_get_irq_optional()
>
> > + if (irq < 0) {
>
> What if this is a real error, or -EPROBE_DEFER?
If it's a real error I believe we should still fallback to polling? Or
do you prefer to only use polling on a fixed condition?
However it's true that I forgot to handle the deferred case here.
> > + dev_info(&pdev->dev, "Using polling\n");
> > + nfc->use_polling = true;
> > + } else {
> > + ret = devm_request_irq(&pdev->dev, irq, rzn1_nfc_irq_handler, 0,
> > + "rzn1-nand-controller", nfc);
> > + if (ret < 0)
> > + return ret;
> > + }
> > +
> > + ret = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
> > + if (ret)
> > + return ret;
> > +
> > + nfc->hclk = devm_clk_get(&pdev->dev, "hclk");
> > + if (IS_ERR(nfc->hclk))
> > + return PTR_ERR(nfc->hclk);
> > +
> > + nfc->eclk = devm_clk_get(&pdev->dev, "eclk");
> > + if (IS_ERR(nfc->eclk))
> > + return PTR_ERR(nfc->eclk);
> > +
> > + ret = clk_prepare_enable(nfc->hclk);
> > + if (ret)
> > + return ret;
> > +
> > + ret = clk_prepare_enable(nfc->eclk);
> > + if (ret)
> > + goto disable_hclk;
> > +
> > + rzn1_nfc_clear_fifo(nfc);
> > +
> > + platform_set_drvdata(pdev, nfc);
> > +
> > + ret = rzn1_nand_chips_init(nfc);
> > + if (ret)
> > + goto disable_eclk;
> > +
> > + return 0;
> > +
> > +disable_eclk:
> > + clk_disable_unprepare(nfc->eclk);
> > +disable_hclk:
> > + clk_disable_unprepare(nfc->hclk);
> > +
> > + return ret;
> > +}
>
> > +static const struct of_device_id rzn1_nfc_id_table[] = {
> > + { .compatible = "renesas,r9a06g032-nand-controller" },
>
> Given my comment on the bindings, you probably want to match against
> "renesas,rzn1-nand-controller" instead.
Sure.
>
> > + {} /* sentinel */
> > +};
> > +MODULE_DEVICE_TABLE(of, nfc_id_table);
> > +
> > +static struct platform_driver rzn1_nfc_driver = {
> > + .driver = {
> > + .name = "renesas-nfc",
>
> Perhaps s/nfc/nandc/ everywhere, to avoid confusion with the other nfc?
There are many NAND controller drivers that are abbreviated with nfc
because it's short and easy to write while still precise, but I have no
issue rewording nfc into nandc if you prefer.
> > + .of_match_table = of_match_ptr(rzn1_nfc_id_table),
> > + },
> > + .probe = rzn1_nfc_probe,
> > + .remove = rzn1_nfc_remove,
> > +};
> > +module_platform_driver(rzn1_nfc_driver);
>
Thanks,
Miquèl
More information about the linux-arm-kernel
mailing list