Micron Confidential > > Are you saying that we may want CQE to stay disabled beyond the FFU > process, to allow the reboot to be completed safely? > Yes, it is. Also there are use cases where reboot is part of the sequence e.g. the force_disable_cmdq flag for the tests is a good example. > > Also we need to claim the host to make sure all the pending commands > > in the queue are completed successfully before disabling. > > Yes, of course. It sounds like you may have misinterpreted my proposals. > > The problem is not that we need to claim the host, but that you add an > entirely new path to do it. > > *If* we conclude that we should add a sysfs node to control CQE, we should > create a mmc blk request for it (which will manage the claim for us as well). I > suggest you have a closer look at power_ro_lock_store(), which should be > the equivalent to what we need to implement here. > I understand what you mean and try to rethink the patch taking inspiration from that. > > > > I can rethink the patch to implement a specific iotcl() request which > > disables CMDQ if you think that is a better implementation but in the > > end it will still require the host claim. > > > > Any feedback or suggestion is appreciated. > > To be clear, I am not proposing to add a new ioctl for mmc. Those we have > today, should be sufficient I think. > > However, rather than adding a new sysfs node to control CQE, perhaps we > can parse the received ioctl data structure, to find out if the > command/request is for FFU and then take specific actions. Along the lines of > what we do for mmc_sanitize(), for example. > > Another option, rather than parsing ioctl data for the FFU command/request, > is to always temporarily disable CQE while processing an mmc ioctl request. > > Would this work? > I think you are right: we should always disable CMDQ for all ioctl requests as of today I do not know a case where CMDQ on is useful for multi cmd lists. What do you suggest: 1 - mimic the power_ro_lock_store and let CMDQ be off across reboots 2 - disable cmdq for multi cmd lists I would prefer 1 because it covers most cases but 2 could be an option I can investigate. > Kind regards > Uffe Cheers! Luca Micron Confidential
