Hi Jennifer,
This is looking better, thanks -- just a few more questions:
On Tue, Nov 16, 2010 at 02:05:16PM +0800, Jennifer Li (TP) wrote:
> Root cause:
> O2Micro's ADMA and related functions can't work by default Linux SD
> driver. We need mark those related registers.
If the goal is to disable ADMA, have you tried using the BROKEN_ADMA
quirk? Like this:
static const struct sdhci_pci_fixes sdhci_o2 = {
.probe = o2_probe,
.quirks = SDHCI_QUIRK_BROKEN_ADMA,
};
If you do this, do you still need to perform the PCI writes?
> + ret = pci_read_config_byte(chip->pdev, O2_SDMMC_MULTI_VCC3V, &scratch);
> + if (ret)
> + return ret;
> +
> + scratch = 0x08;
> + ret = pci_write_config_byte(chip->pdev, O2_SDMMC_MULTI_VCC3V, scratch);
Here you read in a value to scratch, but then you go ahead and write
0x08 to the register regardless. We can skip the read here, right?
> + ret = pci_read_config_byte(chip->pdev, O2_SDMMC_CAPABILITIES, &scratch);
> + if (ret)
> + return ret;
> +
> + scratch |= 0x01;
> + ret = pci_write_config_byte(chip->pdev, O2_SDMMC_CAPABILITIES, scratch);
> +
> + ret = pci_read_config_byte(chip->pdev, O2_SDMMC_CAPABILITIES, &scratch);
> + if (ret)
> + return ret;
> +
> + scratch = 0x73;
> + ret = pci_write_config_byte(chip->pdev, O2_SDMMC_CAPABILITIES, scratch);
Here you write a value to O2_SDMMC_CAPABILITIES that's dependent on the
read value, but then after that you immediately overwrite the register
with 0x73 regardless of what was in it before. Why perform the first
read-and-write cycle and the second read, if you're just going to write
0x73 in the end?
> +
> + ret = pci_read_config_byte(chip->pdev, O2_SDMMC_ADMA1, &scratch);
> + if (ret)
> + return ret;
> +
> + scratch = 0x39;
> + ret = pci_write_config_byte(chip->pdev, O2_SDMMC_ADMA1, scratch);
> +
> + ret = pci_read_config_byte(chip->pdev, O2_SDMMC_ADMA2, &scratch);
> + if (ret)
> + return ret;
> +
> + scratch = 0x08;
> + ret = pci_write_config_byte(chip->pdev, O2_SDMMC_ADMA2, scratch);
Same as above -- if the write isn't dependent on the read, I don't see
why to bother with the read. Also, you can avoid assigning the return
value of pci_write_config_byte() to "ret" now that we aren't testing that.
> Signed-off-by: Jennifer Li Developer <Jennifer.li@xxxxxxxxxxx>
I don't think "Developer" should be here, since the text should just be
your name.
Thanks,
--
Chris Ball <cjb@xxxxxxxxxx> <http://printf.net/>
One Laptop Per Child
--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
