From mboxrd@z Thu Jan  1 00:00:00 1970
From: Stephen Warren <swarren@wwwdotorg.org>
Date: Tue, 05 Feb 2013 16:51:52 -0700
Subject: [U-Boot] [PATCH 2/2] Tegra: MMC: Add DT support to MMC driver
 for all T20 boards
In-Reply-To: <CA+m5__KOLCOKs=5A50FyfPw5Cg4wXrHHeQDx0OMtGCaLzSOCFA@mail.gmail.com>
References: <1360021735-13286-1-git-send-email-twarren@nvidia.com>
	<1360021735-13286-3-git-send-email-twarren@nvidia.com>
	<511165A3.7060003@wwwdotorg.org>
	<CA+m5__KOLCOKs=5A50FyfPw5Cg4wXrHHeQDx0OMtGCaLzSOCFA@mail.gmail.com>
Message-ID: <51119B18.5090807@wwwdotorg.org>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

On 02/05/2013 02:02 PM, Tom Warren wrote:
> Stephen,
> 
> On Tue, Feb 5, 2013 at 1:03 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
>> On 02/04/2013 04:48 PM, Tom Warren wrote:
>>> tegra_mmc_init() now uses DT info for bus width, WP/CD GPIOs, etc.
>>> Tested on Seaboard, fully functional.
>>
>>> diff --git a/board/compal/paz00/paz00.c b/board/compal/paz00/paz00.c
...
>>> @@ -55,18 +55,18 @@ static void pin_mux_mmc(void)
>>>  /* this is a weak define that we are overriding */
>>>  int board_mmc_init(bd_t *bd)
...
>>> +     debug("%s: init eMMC\n", __func__);
>>> +     /* init dev 0, eMMC chip */
>>> +     tegra_mmc_init(0);
...
>>> +     debug("%s: init SD slot\n", __func__);
>>> +     /* init dev 3, SD slot */
>>> +     tegra_mmc_init(3);
>>
>> That doesn't look right. The board code still has knowledge of which
>> SDHCI controllers are in use by the board. Instead, the board should
>> just call tegra_mmc_init() with no parameters at all, and the MMC driver
>> should scan the device tree for all present-and-enabled SDHCI nodes, and
>> instantiate a U-Boot SDHCI device. Without this, the device tree isn't
>> in control of the whole process, so there's little point doing the
>> conversion; a new board couldn't be supported /just/ by creating a new
>> device tree file.
...
>>>  static void tegra_get_setup(struct mmc_host *host, int dev_index)
...
>>> +     int count, node = 0;
>>> +     int node_list[MAX_HOSTS];
>>> +
>>> +     count = fdtdec_find_aliases_for_id(gd->fdt_blob, "sdmmc",
>>> +             COMPAT_NVIDIA_TEGRA20_SDMMC, node_list, MAX_HOSTS);
>>> +     debug("%s: count of nodes is %d\n", __func__, count);
>>> +
>>> +     if (count < dev_index) {
>>> +             printf("%s: device index %d exceeds node count (%d)!\n",
>>> +                     __func__, dev_index, count);
>>> +             return;
>>> +     }
>>
>> This requires that an alias exist in order for the SDHCI node to be
>> found/processed. This isn't correct; the SDHCI nodes must be enumerated
>> themselves, and then the aliases (if any are present) provide a naming
>> hint for them, but even without aliases, the SDHCI nodes must be processed.
>
> Again, I used Allen's SLINK driver for as a template here. In fact, it
> looks like our I2C and SPI/SLINK drivers do this, as well as the
> Exynos SPI and S3C24x0 I2C driver all do this.  Can you point out a
> U-Boot driver that does it the right way (preferably with more than 1
> node, like MMC)? I can take a look at that code to use as an example
> of what you're proposing above.

Tegra's I2C driver has just a single i2c_init_board() function, which
calls fdtdec_find_aliases_for_id() to find all the DT nodes for a given
compatible value (and associated aliases if there are any, I believe),
then calls process_nodes() to loop over them all, and
initialize/register them.