| Message ID | 1363792148-26405-1-git-send-email-thomas.petazzoni@free-electrons.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Mar 20, 2013 at 04:09:08PM +0100, Thomas Petazzoni wrote: > Ralph Droms <rdroms.ietf@gmail.com> reported that 3.9-rc was breaking > the SDIO interface on his Sheevaplug platform, and that the recent > changes to the mvsdio driver are responsible for this breakage. > > After investigation, is turns out that the Sheevaplug does not have > any "card detect" GPIO, and the Sheevaplug has not been converted to > the Device Tree. Therefore, the Sheevaplug board code does not define > a value for the .gpio_card_detect field of the mvsdio_platform_data > structure, which means that its value is 0. Unfortunately, > gpio_is_valid() considers 0 as a valid GPIO, and therefore calls > mmc_gpio_request_cd(), which fails and makes the entire probing of the > driver fail. Hmm, and we have the situation where GPIO 0 is a valid GPIO on some platforms too. So really, we should do something better here. We could either go the route of IRQs and declare GPIO 0 to always be invalid - fixing gpio_is_valid() appropriately, or we should to fix it such that -1 is supplied in the platform data if no GPIOs are specified. However, for -rc I suggest going with your fix which merely restores the old behaviour. > In fact, in the previous mvsdio code, before the Device Tree binding > was introduced, 0 was not considered as a valid GPIO. It would be nice to include the commit reference here where this changed. 07728b77c03d (mmc: mvsdio: use slot-gpio for card detect gpio). -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Russell, On Wed, 20 Mar 2013 17:40:43 +0000, Russell King - ARM Linux wrote: > > After investigation, is turns out that the Sheevaplug does not have > > any "card detect" GPIO, and the Sheevaplug has not been converted to > > the Device Tree. Therefore, the Sheevaplug board code does not define > > a value for the .gpio_card_detect field of the mvsdio_platform_data > > structure, which means that its value is 0. Unfortunately, > > gpio_is_valid() considers 0 as a valid GPIO, and therefore calls > > mmc_gpio_request_cd(), which fails and makes the entire probing of the > > driver fail. > > Hmm, and we have the situation where GPIO 0 is a valid GPIO on some > platforms too. So really, we should do something better here. Agreed. > We could either go the route of IRQs and declare GPIO 0 to always be > invalid - fixing gpio_is_valid() appropriately, or we should to fix > it such that -1 is supplied in the platform data if no GPIOs are > specified. I haven't followed all the details, but isn't the gpiod work also a way of solving this problem? Also, the platform data stuff is going away, at least for the users of the mvsdio driver, since they are being converted to the Device Tree, and of_get_gpio() properly returns an error code when no GPIO has been specified in the DT. > However, for -rc I suggest going with your fix which merely restores > the old behaviour. Yes > > In fact, in the previous mvsdio code, before the Device Tree binding > > was introduced, 0 was not considered as a valid GPIO. > > It would be nice to include the commit reference here where this changed. > 07728b77c03d (mmc: mvsdio: use slot-gpio for card detect gpio). Sure, I'll resend a v3 that includes this. Thanks for the review! Thomas
diff --git a/drivers/mmc/host/mvsdio.c b/drivers/mmc/host/mvsdio.c index 145cdaf..1e4d567 100644 --- a/drivers/mmc/host/mvsdio.c +++ b/drivers/mmc/host/mvsdio.c @@ -741,8 +741,8 @@ static int __init mvsd_probe(struct platform_device *pdev) goto out; } host->base_clock = mvsd_data->clock / 2; - gpio_card_detect = mvsd_data->gpio_card_detect; - gpio_write_protect = mvsd_data->gpio_write_protect; + gpio_card_detect = mvsd_data->gpio_card_detect ? : -EINVAL; + gpio_write_protect = mvsd_data->gpio_write_protect ? : -EINVAL; } mmc->ops = &mvsd_ops;
Ralph Droms <rdroms.ietf@gmail.com> reported that 3.9-rc was breaking the SDIO interface on his Sheevaplug platform, and that the recent changes to the mvsdio driver are responsible for this breakage. After investigation, is turns out that the Sheevaplug does not have any "card detect" GPIO, and the Sheevaplug has not been converted to the Device Tree. Therefore, the Sheevaplug board code does not define a value for the .gpio_card_detect field of the mvsdio_platform_data structure, which means that its value is 0. Unfortunately, gpio_is_valid() considers 0 as a valid GPIO, and therefore calls mmc_gpio_request_cd(), which fails and makes the entire probing of the driver fail. In fact, in the previous mvsdio code, before the Device Tree binding was introduced, 0 was not considered as a valid GPIO. Therefore, this fix revert back to this behavior in the non-DT case, by setting the gpio_card_detect and gpio_write_protect local variables to -EINVAL when the corresponding fields of the mvsdio_platform_data structure are set to zero (i.e, left undefined). Of course, it prevents to use GPIO 0 as a card detect or write protect GPIO, but it was a defiency of the previous non-DT code, and the fix moving forward is to convert platforms to the Device Tree. The problem has been reproduced successfully on the Kirkwood-based Marvell DB-88F6281-BP Development Board (that doesn't use the Device Tree) and the fix has proven to work properly, after of course removing the gpio_card_detect field of the mvsdio_platform_data instance for this board. Reported-by: Ralph Droms <rdroms.ietf@gmail.com> Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com> --- This patch should be applied on 3.9-rcX. Changes between v1 and v2: * Send the patch to the MMC maintainer instead of the Marvell maintainers. --- drivers/mmc/host/mvsdio.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)