Am Thu, 19 Sep 2024 13:02:55 +0200 schrieb Krzysztof Kozlowski <krzk@xxxxxxxxxx>: > On 19/09/2024 12:50, Andreas Kemnade wrote: > > Am Wed, 18 Sep 2024 15:43:40 -0700 > > schrieb Guenter Roeck <linux@xxxxxxxxxxxx>: > > > >> On 9/18/24 14:29, Andreas Kemnade wrote: > >>> Avoid requiring MODULE_ALIASES by declaring proper device id > >>> tables. > >>> > >>> Signed-off-by: Andreas Kemnade <andreas@xxxxxxxxxxxx> > >> > >> This needs a better rationale. There are more than 40 watchdog > >> drivers using MODULE_ALIAS. I would hate having to deal with 40+ > >> patches just for cosmetic reasons, not counting the thousands of > >> instances of MODULE_ALIAS in the kernel, including the more than > >> 1,000 instances of "MODULE_ALIAS.*platform:". > >> > > basically reviewers were arguing against patches from me bringing in > > MODULE_ALIASES. So I decided to clean up a bit in my backyard. Not > > sure whether such things could by done by coccinelle but at least > > it could be tested via output of modinfo. > > > > This is one example for such a patch: > > https://lore.kernel.org/linux-clk/119f56c8-5f38-eb48-7157-6033932f0430@xxxxxxxxxx/ > > > > There are multiple aspects here: > 1. People (including me) copy code which they do no understand. Or > without really digging into it, because they do not have time. They > just copy it, regardless whether the code is necessary or not. > MODULE_ALIAS is one of such examples. It got copied to new drivers > just because some other driver had it. > and copy nowadays unaccepted design patterns. Probably best to look at the newest example. > 2. MODULE_ALIAS creates basically ABI - some user-space might depend > on it, so removal might affect user. I think I was not dropping it > from the drivers in cases it would actually drop an alias. I was only > dropping duplicated aliases. That's not the case here, I believe. > > 3. MODULE_ALIAS scales poor. I believe proper xxx_device_id table is > better. > > 4. But it does not mean that one single line - MODULE_ALIAS - should > be replaced in existing drivers into full-blown ID table. I think I > never proposed such patches for existing drivers. Why? Because if > there was no such need so far, means there were no scalability issues. > Thanks for the long explanation. Regards, Andreas
