On Sat, Jan 8, 2022 at 4:32 AM Gabriel L. Somlo <gsomlo@xxxxxxxxx> wrote: > On Fri, Jan 07, 2022 at 08:57:43PM -0500, Gabriel L. Somlo wrote: > > On Sat, Jan 08, 2022 at 01:54:07AM +0200, Andy Shevchenko wrote: > > > On Saturday, January 8, 2022, Gabriel Somlo <gsomlo@xxxxxxxxx> wrote: ... > > > + mmc = mmc_alloc_host(sizeof(struct litex_mmc_host), &pdev->dev); > > > > > > Should be devm or you may not use devm at all. See hint in one of the previous > > > messages in v7 discussion. > > > > And here I think I'm in trouble... :) > > > > None of the examples retrieved via > > > > `git log --no-merges --grep devm_add_action_or_reset` > > > > are from "drivers/mmc/host/*", and *all* of the mmc drivers there, > > including the ones that make extensive use of devm_*, use > > mmc_alloc_host(), and there doesn't appear to be a devm-ified version > > of mmc_alloc_host() available! How do they all get away with it? > > > > I'm really confused now -- any additional clue(s) much appreciated! > > I found drivers/mmc/host/meson-mx-sdhc-mmc.c, which uses > devm_add_action_or_reset() right after mmc_alloc_host() to enlist the > subsequent call to mmc_free_host(), see here: > > https://github.com/torvalds/linux/blob/master/drivers/mmc/host/meson-mx-sdhc-mmc.c#L791 > > This would mean that I no longer have to call mmc_free_host(), neither > on probe()'s error path, nor during remove(). > > Does that count as canonically correct, or am I still missing > something? Yes, this is one of the options you may use. Since it will be a second (?) driver with the same idea, perhaps in the future it would make sense to provide devm_mmc_alloc_host() or analogue (in the latter it means more complex solution like the input subsystem is using, see devm_input_... API implementations). -- With Best Regards, Andy Shevchenko
