Re: [PATCH 2/3] mtd: spi-nor: core code for the Altera Quadspi Flash Controller v2

[Marek Vasut <marek.vasut@gmail.com>]

On 06/27/2017 04:57 PM, matthew.gerlach@linux.intel.com wrote:

[...]

>>> +static int altera_quadspi_read_reg(struct spi_nor *nor, u8 opcode,
>>> u8 *val,
>>> +                   int len)
>>> +{
>>> +    struct altera_quadspi_flash *flash = nor->priv;
>>> +    struct altera_quadspi *q = flash->q;
>>> +    u32 data = 0;
>>> +
>>> +    memset(val, 0, len);
>>> +
>>> +    altera_quadspi_chip_select(q, flash->bank);
>>> +
>>> +    switch (opcode) {
>>> +    case SPINOR_OP_RDSR:
>>> +        data = alt_qspi_readl(q->csr_base, QUADSPI_SR_REG);
>>> +        dev_dbg(q->dev, "%s RDSR 0x%x\n", __func__, data);
>>> +        *val = (u8)data & QUADSPI_SR_MASK;
>>> +        break;
>>> +    case SPINOR_OP_RDID:
>>> +        if (q->opcode_id == EPCS_OPCODE_ID)
>>> +            data = alt_qspi_readl(q->csr_base, QUADSPI_SID_REG);
>>> +        else
>>> +            data = alt_qspi_readl(q->csr_base, QUADSPI_RDID_REG);
>>> +
>>> +        *((u32 *)val) = data;
>>
>> What are these awful casts ?
> 
> This component requires reading the registers as 32 bit quantities. So it
> seemed the right thing to do to me.

Does this handle endianness well ?

>>> +        break;
>>> +    case SPINOR_OP_RDFSR:
>>> +        data = alt_qspi_readl(q->csr_base, QUADSPI_FLAG_STATUS_REG);
>>> +        dev_dbg(q->dev, "%s RDFSR 0x%x\n", __func__, data);
>>> +        *val = (u8)(data & 0xff);
>>> +        break;
>>> +    default:
>>> +        dev_dbg(q->dev, "%s UNHANDLED read_reg 0x%x\n",
>>> +            __func__, opcode);
>>> +        *val = 0;
>>> +        break;
>>> +    }
>>> +    return 0;
>>> +}

[...]

>>> +#define WINDOW_ALIGN 4
>>> +#define WINDOW_MASK (WINDOW_ALIGN - 1)
>>
>> What are these undocumented macros in the middle of the code ?
> 
> The bindings document states that when a windowed bridge
> is used, all accesses must be 32 bit.  I can comment/rename
> and put at the top.

Yes please.

[...]

>>> diff --git a/include/linux/mtd/altera-quadspi.h
>>> b/include/linux/mtd/altera-quadspi.h
>>> new file mode 100644
>>> index 0000000..58f31ee
>>> --- /dev/null
>>> +++ b/include/linux/mtd/altera-quadspi.h
>>> @@ -0,0 +1,28 @@
>>> +/*
>>> + *
>>> + * Copyright 2017 Intel Corporation, Inc.
>>> + *
>>> + * This program is free software; you can redistribute it and/or modify
>>> + * it under the terms of the GNU General Public License as published by
>>> + * the Free Software Foundation; either version 2 of the License, or
>>> + * (at your option) any later version.
>>> + */
>>> +#ifndef __ALTERA_QUADSPI_H
>>> +#define __ALTERA_QUADSPI_H
>>> +
>>> +#include <linux/device.h>
>>> +
>>> +#define ALTERA_QUADSPI_FL_BITREV_READ BIT(0)
>>> +#define ALTERA_QUADSPI_FL_BITREV_WRITE BIT(1)
>>> +
>>> +#define ALTERA_QUADSPI_MAX_NUM_FLASH_CHIP 3
>>> +
>>> +int altera_quadspi_create(struct device *dev, void __iomem *csr_base,
>>> +              void __iomem *data_base, void __iomem *window_reg,
>>> +              size_t window_size, u32 flags);
>>> +
>>> +int altera_qspi_add_bank(struct device *dev,
>>> +             u32 bank, struct device_node *np);
>>> +
>>> +int altera_quadspi_remove_banks(struct device *dev);
>>
>> Why is this header needed at all ?
> 
> This header is needed because of the very different ways
> FPGAs can be used with a processor running Linux.  In the case of a
> soft processor in the FPGA or an ARM connected to a FPGA, this header
> is not necessary because device trees are used to probe the driver.
> However, if the FPGA is on a PCIe card connected to an x86, device trees
> are not generally used, and the pcie driver must enumerate the
> "sub-driver".

But we don't support that later part, do we ?

-- 
Best regards,
Marek Vasut