On Fri, May 22, 2020 at 07:10:01PM +0300, Serge Semin wrote: > On Fri, May 22, 2020 at 07:04:17PM +0300, Andy Shevchenko wrote: > > We may improve it later, though, to be smarter and spill a warning only when it > > uses non-atomic delays. For now this is good enough. > IMO this wouldn't be better than the current solution. might_sleep() is called > "might" to warn that the called method may get to sleep, not shall, not will. As > I see it it's better to warn about the consequences straight away, but not at the > point when the sleeping method is actually called in the atomic context. Yes, this is the whole point of might_sleep() - it's to ensure that we always get a warning about code that might sleep, rather than only getting a warning if we happen to go down a path that does actually sleep. This means we're less likely to run into bugs at runtime in unusual use cases.
Attachment:
signature.asc
Description: PGP signature
