On Mon, Nov 23, 2009 at 11:15 PM, Tony Lindgren <tony@xxxxxxxxxxx> wrote:
> * Vimal Singh <vimal.newwork@xxxxxxxxx> [091119 00:52]:
>> On Wed, Nov 18, 2009 at 10:37 PM, Tony Lindgren <tony@xxxxxxxxxxx> wrote:
>> > * Vimal Singh <vimal.newwork@xxxxxxxxx> [091118 06:38]:
>> >> Tony,
>> >>
>> >> On Fri, Nov 13, 2009 at 2:14 AM, Tony Lindgren <tony@xxxxxxxxxxx> wrote:
>> >> > * Vimal Singh <vimal.newwork@xxxxxxxxx> [091110 02:08]:
>> >> >> From 6f535d7128ca392458dd0cb31d138cda84747c06 Mon Sep 17 00:00:00 2001
>> >> >> From: Vimal Singh <vimalsingh@xxxxxx>
>> >> >> Date: Tue, 10 Nov 2009 11:42:45 +0530
>> >> >> Subject: [PATCH] OMAP3: Add support for NAND on ZOOM2/LDP boards
>> >>
>> >> [...]
>> >>
>> >> >> +static int omap_ldp_nand_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
>> >> >> +{
>> >> >> + int ret = 0;
>> >> >> + int chipnr;
>> >> >> + int status;
>> >> >> + unsigned long page;
>> >> >> + struct nand_chip *this = mtd->priv;
>> >> >> + printk(KERN_INFO "nand_unlock: start: %08x, length: %d!\n",
>> >> >> + (int)ofs, (int)len);
>> >> >> +
>> >> >> + /* select the NAND device */
>> >> >> + chipnr = (int)(ofs >> this->chip_shift);
>> >> >> + this->select_chip(mtd, chipnr);
>> >> >> + /* check the WP bit */
>> >> >> + this->cmdfunc(mtd, NAND_CMD_STATUS, -1, -1);
>> >> >> + if ((this->read_byte(mtd) & 0x80) == 0) {
>> >> >> + printk(KERN_ERR "nand_unlock: Device is write protected!\n");
>> >> >> + ret = -EINVAL;
>> >> >> + goto out;
>> >> >> + }
>> >> >> +
>> >> >> + if ((ofs & (mtd->writesize - 1)) != 0) {
>> >> >> + printk(KERN_ERR "nand_unlock: Start address must be"
>> >> >> + "beginning of nand page!\n");
>> >> >> + ret = -EINVAL;
>> >> >> + goto out;
>> >> >> + }
>> >> >> +
>> >> >> + if (len == 0 || (len & (mtd->writesize - 1)) != 0) {
>> >> >> + printk(KERN_ERR "nand_unlock: Length must be a multiple of "
>> >> >> + "nand page size!\n");
>> >> >> + ret = -EINVAL;
>> >> >> + goto out;
>> >> >> + }
>> >> >> +
>> >> >> + /* submit address of first page to unlock */
>> >> >> + page = (unsigned long)(ofs >> this->page_shift);
>> >> >> + this->cmdfunc(mtd, NAND_CMD_UNLOCK1, -1, page & this->pagemask);
>> >> >> +
>> >> >> + /* submit ADDRESS of LAST page to unlock */
>> >> >> + page += (unsigned long)((ofs + len) >> this->page_shift) ;
>> >> >> + this->cmdfunc(mtd, NAND_CMD_UNLOCK2, -1, page & this->pagemask);
>> >> >> +
>> >> >> + /* call wait ready function */
>> >> >> + status = this->waitfunc(mtd, this);
>> >> >> + udelay(1000);
>> >> >> + /* see if device thinks it succeeded */
>> >> >> + if (status & 0x01) {
>> >> >> + /* there was an error */
>> >> >> + printk(KERN_ERR "nand_unlock: error status =0x%08x ", status);
>> >> >> + ret = -EIO;
>> >> >> + goto out;
>> >> >> + }
>> >> >> +
>> >> >> + out:
>> >> >> + /* de-select the NAND device */
>> >> >> + this->select_chip(mtd, -1);
>> >> >> + return ret;
>> >> >> +}
>> >> >
>> >> > Isn't the unlocking generic to the NAND device driver? Or is it somehow board
>> >> > specific?
>> >>
>> >> Yes, zoom2 boards come up with whole NAND locked, whereas it is not
>> >> case for SDP boards.
>> >
>> > But the procedure should be done under drivers/mtd I believe using some
>> > standard tools.
>>
>> OK, I'll take this discussion to mtd mailing list. For now I'll remove
>> this from here.
>
> OK, I'd assume there's some standard way to handle this for all NOR
> drivers.
>
>> >
>> >> >
>> >> >
>> >> >> +static struct mtd_partition ldp_nand_partitions[] = {
>> >> >> + /* All the partition sizes are listed in terms of NAND block size */
>> >> >> + {
>> >> >> + .name = "X-Loader-NAND",
>> >> >> + .offset = 0,
>> >> >> + .size = 4 * (64 * 2048), /* 512KB, 0x80000 */
>> >> >> + .mask_flags = MTD_WRITEABLE, /* force read-only */
>> >> >> + },
>> >> >> + {
>> >> >> + .name = "U-Boot-NAND",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x80000 */
>> >> >> + .size = 10 * (64 * 2048), /* 1.25MB, 0x140000 */
>> >> >> + .mask_flags = MTD_WRITEABLE, /* force read-only */
>> >> >> + },
>> >> >> + {
>> >> >> + .name = "Boot Env-NAND",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x1c0000 */
>> >> >> + .size = 2 * (64 * 2048), /* 256KB, 0x40000 */
>> >> >> + },
>> >> >> + {
>> >> >> + .name = "Kernel-NAND",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x0200000*/
>> >> >> + .size = 240 * (64 * 2048), /* 30M, 0x1E00000 */
>> >> >> + },
>> >> >> +#ifdef CONFIG_MACH_OMAP_ZOOM2
>> >> >> + {
>> >> >> + .name = "system",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x2000000 */
>> >> >> + .size = 1280 * (64 * 2048), /* 160M, 0xA000000 */
>> >> >> + },
>> >> >> + {
>> >> >> + .name = "userdata",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0xC000000 */
>> >> >> + .size = 256 * (64 * 2048), /* 32M, 0x2000000 */
>> >> >> + },
>> >> >> + {
>> >> >> + .name = "cache",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0xE000000 */
>> >> >> + .size = 256 * (64 * 2048), /* 32M, 0x2000000 */
>> >> >> + },
>> >> >> +#else
>> >> >> + {
>> >> >> + .name = "File System - NAND",
>> >> >> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x2000000 */
>> >> >> + .size = MTDPART_SIZ_FULL, /* 96MB, 0x6000000 */
>> >> >> + },
>> >> >> +#endif
>> >> >
>> >> > Please remove the ifdefs, you should be able to compile in all mach-omap2
>> >> > boards into the same kernel binary. You should set the size dynamically as
>> >> > needed.
>> >>
>> >> We can always provide partitioning information from cmdline
>> >> (mtdparts:). Which will anyway have higher precedence than this. Here
>> >> are trying provide default static partitions. Which IMHO is fine.
>> >>
>> >> see this too:
>> >> http://marc.info/?l=linux-omap&m=125178914327011&w=2
>> >
>> > No way we're adding any more of these ifdef else hacks.
>> >
>> > The whole idea of the platform_data is to pass the board specific
>> > configuration to the driver. The driver should be generic.
>> >
>>
>> Hmmm... In that case I'll remove extra partition informations (for
>> zoom2 and sdp) and will have very only few common partitions here.
>> Something like this:
>>
>> +static struct mtd_partition ldp_nand_partitions[] = {
>> + /* All the partition sizes are listed in terms of NAND block size */
>> + {
>> + .name = "X-Loader-NAND",
>> + .offset = 0,
>> + .size = 4 * (64 * 2048), /* 512KB, 0x80000 */
>> + .mask_flags = MTD_WRITEABLE, /* force read-only */
>> + },
>> + {
>> + .name = "U-Boot-NAND",
>> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x80000 */
>> + .size = 10 * (64 * 2048), /* 1.25MB, 0x140000 */
>> + .mask_flags = MTD_WRITEABLE, /* force read-only */
>> + },
>> + {
>> + .name = "Boot Env-NAND",
>> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x1c0000 */
>> + .size = 2 * (64 * 2048), /* 256KB, 0x40000 */
>> + },
>> + {
>> + .name = "Kernel-NAND",
>> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x0200000*/
>> + .size = 240 * (64 * 2048), /* 30M, 0x1E00000 */
>> + },
>> + {
>> + .name = "File System - NAND",
>> + .offset = MTDPART_OFS_APPEND, /* Offset = 0x2000000 */
>> + .size = MTDPART_SIZ_FULL,
>> + },
>> +};
>>
>>
>> And then, as I said earlier, if someone want different partitions, he
>> can always pass his partition information from cmdline (bootargs).
>
> Why don't you just set ldp_nand_paritions[X].size = something during init
> based on the machine_is_XXX()?
>
>> >
>> >>
>> >> >
>> >> >
>> >> >> +};
>> >> >> +
>> >> >> +/* NAND chip access: 16 bit */
>> >> >> +static struct omap_nand_platform_data ldp_nand_data = {
>> >> >> + .parts = ldp_nand_partitions,
>> >> >> + .nr_parts = ARRAY_SIZE(ldp_nand_partitions),
>> >> >> + .nand_setup = NULL,
>> >> >> + .dma_channel = -1, /* disable DMA in OMAP NAND driver */
>> >> >> + .dev_ready = NULL,
>> >> >> + .unlock = omap_ldp_nand_unlock,
>> >> >> +};
>> >> >> +
>> >> >> +static struct resource ldp_nand_resource = {
>> >> >> + .flags = IORESOURCE_MEM,
>> >> >> +};
>> >> >> +
>> >> >> +static struct platform_device ldp_nand_device = {
>> >> >> + .name = "omap2-nand",
>> >> >> + .id = 0,
>> >> >> + .dev = {
>> >> >> + .platform_data = &ldp_nand_data,
>> >> >> + },
>> >> >> + .num_resources = 1,
>> >> >> + .resource = &ldp_nand_resource,
>> >> >> +};
>> >> >> +
>> >> >> +/**
>> >> >> + * ldp430_flash_init - Identify devices connected to GPMC and register.
>> >> >> + *
>> >> >> + * @return - void.
>> >> >> + */
>> >> >> +void __init zoom_flash_init(void)
>> >> >> +{
>> >> >> + u8 nandcs = GPMC_CS_NUM + 1;
>> >> >> + u32 gpmc_base_add = OMAP34XX_GPMC_VIRT;
>> >> >> +
>> >> >> + /* pop nand part */
>> >> >> + nandcs = LDP3430_NAND_CS;
>> >> >> +
>> >> >> + ldp_nand_data.cs = nandcs;
>> >> >> + ldp_nand_data.gpmc_cs_baseaddr = (void *)(gpmc_base_add +
>> >> >> + GPMC_CS0_BASE + nandcs * GPMC_CS_SIZE);
>> >> >> + ldp_nand_data.gpmc_baseaddr = (void *) (gpmc_base_add);
>> >> >> +
>> >> >> + if (platform_device_register(&ldp_nand_device) < 0)
>> >> >> + printk(KERN_ERR "Unable to register NAND device\n");
>> >> >> +}
>> >> >
>> >> > This too should use gpmc_cs_request().
>> >>
>> >> This is just platform data, needed by nand driver (nand/omap2.c).
>> >> 'gpmc_cs_request' is separately done in mentioned driver. And IMO that
>> >> is the correct place to do this, as every time (when driver is
>> >> compiled as module) inserting and de-inserting driver, will need this.
>> >
>> > Without gpmc_cs_request the GPMC can be in uninitialized state.
>> > You're basically relying on some hardcoded values from the bootloader,
>> > which can be at whatever version and whatever bootloader. Not good.
>>
>> As I said, 'gpmc_cs_request' is done in the mentioned driver itself.
>> So, GPMC (cs) gets initialized at that point. And this is exactly why
>> I said "calling 'gpmc_cs_request' from driver is fine. Because every
>> time you remove driver module (rmmod), you expect 'cs' for that driver
>> gets disable (done in gpmc_cs_free) and if insert the driver again it
>> gets enabled again (done in gpmc_cs_request).
>
> Well you should be able to use totally standard MTD drivers
> by doing only the omap specific initialization under mach-omap2.
>
> For any new drivers, calling gpmc from the driver itself just seems
> wrong. That's something we want to stop doing, not add more of it.
>
I still think 'gpmc_cs_request' should be done from driver itself, why
to get a resource allocated if driver itself is not present (in case
driver is compile as a module and yet not inserted). As it is already
done in 'onenand/omap2.c' and 'nand/omap2.c'.
Though, for NOR we can do 'gpmc_cs_request' in 'board-*-flash.c'
itself. As I did in my patch series V6.
--
Regards,
Vimal Singh
> How about instead just set up a generic mach-omap2/gpmc-nor.c
> and use some standard MTD driver?
>
> Regards,
>
> Tony
>
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
