Hi Thomas/Simon,
On 4 November 2015 at 08:42, Thomas Chou thomas@wytron.com.tw wrote:
On 2015年11月03日 22:55, Jagan Teki wrote:
On 3 November 2015 at 20:19, Thomas Chou thomas@wytron.com.tw wrote:
Hi Jagan,
On 2015年11月03日 22:41, Jagan Teki wrote:
Hi Thomas,
On 3 November 2015 at 18:39, Thomas Chou thomas@wytron.com.tw wrote:
Implement a Memory Technology Device (MTD) uclass. It should include most flash drivers in the future. Though no uclass ops are defined yet, the MTD ops could be used.
The NAND flash driver is based on MTD. The CFI flash and SPI flash support MTD, too. It should make sense to convert them to MTD uclass.
Why does MTD require driver model? Should drivers like nand, cfi or etc register mtd core should need to move on dm?
The driver model combined with device tree control of u-boot offers dynamic binding of drivers and devices. It is expected that all drivers will be converted to driver model, including nand, cfi and spi flash.
So, mtd_info ops like _erase, _write and _read will also change or something like this
struct dm_mtd_info { struct mtd_info *info; struct udevice *dev; };
Not exactly. I included udevice in mtd_info as it was device for Linux.
@@ -272,6 +273,8 @@ struct mtd_info { struct module *owner; #ifndef __UBOOT__ struct device dev; +#else
struct udevice *dev;#endif int usecount; };
I think the mtd ops is more complete and widely used. There might be no need to reinvent the dm_mtd ops. The mtd uclass priv is set to mtd_info and we can get it with mtd_get_info(dev). Then call mtd ops, like mtd_read() mtd_write and mtd_erase(), directly.
See for example, I have recently added MTD support to spi_flash [1] [2]
[1] https://patchwork.ozlabs.org/patch/529397/ [2] https://patchwork.ozlabs.org/patch/529399/
It seems we are working toward the same direction. :)
Sorry, I couldn't understand this looks we're in different direction.
Let me explain what I thought about mtd_info usage. udevice should be part of underlying flash structure's like cfi, nand and spi_flash and mtd_info should be used for it's core api's like _erase. _read and _write and underlying driver will use their global structure that include's mtd and udevice as a function pointer like this.
struct spi_flash { struct mtd_info *info; struct udevice *device; }
struct spi_flash_priv { struct spi_flash flash; struct mtd_info mtd; };
static int spi_flash_std_probe(struct udevice *dev) { struct spi_flash_priv *priv = dev_get_uclass_priv(dev); struct spi_slave *spi = dev_get_parent_priv(dev); struct spi_flash *flash; int ret;
flash = &priv->flash; flash->mtd = &priv->mtd;
flash->spi = spi; flash->priv = priv;
priv->mtd.priv = flash; flash->dev = dev; }
U_BOOT_DRIVER(spi_flash_std) = { .name = "spi_flash_std", .id = UCLASS_SPI_FLASH, .of_match = spi_flash_std_ids, .probe = spi_flash_std_probe, .priv_auto_alloc_size = sizeof(struct spi_flash_priv), };
This is the way I have implemented mtd on spi-flash[1] [2] [1] https://patchwork.ozlabs.org/patch/529397/ [2] https://patchwork.ozlabs.org/patch/529399/
Please explain how this related your approach of adding udevice to mtd.
Simon suggested that we can have an unified flash class (for all cfi, spi and nand flash) after the discussion between Bin Meng and I. So I dropped the earlier cfi-flash uclass, and found the mtd might be a better uclass. We see the same point, "MTD has proven core for flash operations".
The work on cfi-flash is not complete yet. It needs to reshape to use mtd ops like your earlier patches. But I have to work on others.
The spi-flash uclass should be merged into mtd uclass and use mtd ops. Maybe you will be interested and will help. Thanks in advance.
The nand flash is more ready. But need to convert to driver model.
thanks!