On 16/04/15 18:19, Ulf Hansson wrote: > [...] > >>>>> 1) Card removal/detect (hold/release?) >>>> >>>> Re-tuning will be done if needed before detect (because it is done before a >>>> request), which is necessary because detect can fail if tuning is needed. >>>> >>>> Re-tuning is done before a request. Requests aren't started if the card has >>>> been removed i.e. mmc_start_request() calls mmc_card_removed() >>>> >>>> If tuning is executed while a card is being removed, then the tuning will >>>> fail and the request will be errored out. >>> >>> So you are saying that we instead of relying on the CMD13 (for SD/MMC) >>> to check whether the card is still alive, it's fine to trigger a >>> re-tuning instead. >> >> (Oops forgot to answer this one, sorry) >> >> Yes, why not? >> >>> >>> I don't think that is an effective way to remove a card. >> >> Generally we know the card is removed from card-detect so no commands are >> sent in either case. > > Not sure what you mean here. The sdhci driver won't issue commands if card-detect shows no card. It just sets the error to -ENOMEDIUM. > > In case when the card is "idle" and the host sees a GPIO CD irq, it > will trigger a detect work to run mmc_rescan(). > > In this case, it's the responsibility of mmc_rescan() to find out if > the card is being removed. It does so by invoking the > bus_ops->detect() callback, which eventually will send a CMD13 for > mmc/sd. > > In this scenario, I can't see why we want to allow re-tuning to happen > in front of the CMD13 command. If re-tuning is needed and can be done, it is done. It doesn't need to be more complicated than that. > >> >>> >>> Moreover, I find it quite unreasonable to say the check for an alive >>> card, would fail because of that a re-tuning is needed. That would in >>> principle mean that we never should be able to hold re-tune for any >>> commands sequences. >> >> Re-tuning must work if the card is alive. CMD13 might get a CRC error if >> re-tuning is needed. > > And that then applies to all commands which we hold re-tuning for. So > then we can't _ever_ hold re-tuning for any command, since we might > get CRC errors. Periodic re-tuning and re-tuning after pm-runtime-resume should ensure no CRC errors while re-tuning is held. -- 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
