On 22 November 2016 at 21:56, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote: > Changes in v2: > > - From discussions around the difficulties for how to support CMD13 > polling for HS200, HS400 HS400ES, I decided to drop those patches for > now. It's is particularly due to the need for tuning, after a speed mode > switch, that makes it hard to rely on CMD13 polling for these cases. > Perhaps we can allow CMD13 polling when switching to the intermediate > speed modes, but let's deal with that outside of this series. > > - Folded in a new patch, which checks the SWITCH_ERROR bit in each > CMD13 response when polling with CMD13. Patch v2 4/7. > > - No other changes made to the rest of the series. > > > > Several changes has been made for how and when to use CMD13 as a > polling-method, to find out when the mmc cards stops signaling busy after > sending a CMD6. This particularly concerns the cases when switching to new bus > speed modes, such as HS (high-speed), HS DDR, HS200, HS400 and HS400ES. > > Currently CMD13 polling is *not* allowed for these cases, but this was recently > changed - and which did cause regressions for card detection for some > platforms. > > A simple fix was made to solve these regressions, simply by using a default > CMD6 generic timeout of 500ms, as to avoid using the fall-back timeout in > __mmc_switch() of 10min. That greatly improve the situation and one could > consider the regressions as "solved". > > However, this simple fix is still causing unnecessary long card initialization > times, especially for those mmc hosts that doesn't support HW busy detection > (using MMC_CAP_WAIT_WHILE_BUSY and/or implements the ->card_busy() host ops). > > This because we wait a fixed amount of time (CMD6 generic timeout) to make sure > the card is done signaling busy, while in practice this happens a lot sooner. > > A couple of tests for HS and HS DDR eMMC cards, shows the card only need a few > ms to complete the bus speed mode changes, thus it's far less than the CMD6 > generic timeout. > > To improve this behaviour and shorten the card initialization time, we need to > allow using CMD13 as polling method to find out when the card stops signaling > busy. Although, enabling CMD13 polling may also introduce other issues as > according to the JEDEC spec, it's not the recommended method. Especially it > mentions that CRC errors may be triggered when sending a CMD13 command during a > bus timing change. > > To deal with these issues, we need to change from ignoring those CRC errors but > instead continue to treat the card as being busy and continue to poll with > CMD13. Perhaps this behaviour is one of reasons to why the earlier CMD13 polling > method didn't work out well. > > This series requires extensive testing, please help with that! > > I have currently tested it with HS and HS DDR eMMC cards, and for combinations > of an mmc hosts (mmci) supporting HW busy detection and not (local hacks in > mmci.c). > > > Ulf Hansson (9): > mmc: core: Retry instead of ignore at CRC errors when polling for busy > mmc: core: Remove redundant __mmc_send_status() > mmc: core: Rename ignore_crc to retry_crc_err to reflect its purpose > mmc: core: Enable __mmc_switch() to change bus speed timing for the > host > mmc: core: Allow CMD13 polling when switching to HS mode for mmc > mmc: core: Update CMD13 polling policy when switch to HS DDR mode > mmc: core: Allow CMD13 polling when switch to HS200 mode > mmc: core: Allow CMD13 polling when switch to HS400 mode > mmc: core: Allow CMD13 polling when switch to HS400ES mode > > drivers/mmc/core/core.c | 2 +- > drivers/mmc/core/mmc.c | 156 ++++++++++++++------------------------------- > drivers/mmc/core/mmc_ops.c | 45 +++++++------ > drivers/mmc/core/mmc_ops.h | 4 +- > 4 files changed, 77 insertions(+), 130 deletions(-) Oops, forgot to clean up the cover letter. Please ignore the summary of changes above. It's the below changes that are the v2 series. Apologize for the inconvenience! Kind regards Uffe > > -- > 1.9.1 > > > > > > > *** BLURB HERE *** > > Ulf Hansson (7): > mmc: core: Retry instead of ignore at CRC errors when polling for busy > mmc: core: Remove redundant __mmc_send_status() > mmc: core: Rename ignore_crc to retry_crc_err to reflect its purpose > mmc: core: Check SWITCH_ERROR bit from each CMD13 response when > polling > mmc: core: Enable __mmc_switch() to change bus speed timing for the > host > mmc: core: Allow CMD13 polling when switching to HS mode for mmc > mmc: core: Update CMD13 polling policy when switch to HS DDR mode > > drivers/mmc/core/core.c | 2 +- > drivers/mmc/core/mmc.c | 53 +++++++++++++++++++++++----------------------- > drivers/mmc/core/mmc_ops.c | 51 ++++++++++++++++++++++++++------------------ > drivers/mmc/core/mmc_ops.h | 4 ++-- > 4 files changed, 60 insertions(+), 50 deletions(-) > > -- > 1.9.1 > -- 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
