Hello, Manolo de Medici, le lun. 06 avril 2026 00:02:33 +0100, a ecrit: > This defines just the API, nothing is changed in the scheduler. I'd like to > get feedback on the choices I made. > > Unfortunately, since in GNU/Hurd thread creation pass through the Task API, > and not the thread API, affinity inheritance will have to be implemented in > glibc. > > Difference between v2 and v1: now allows to specify affinity for processors > outside of the pset, to mimick Linux behavior more closely. I am a bit > worried that the removal of all cpus in the affinity set would cause a > process to freeze. TODO: either reset all affinities of all thread in set > on processor removal, or assure that at least one cpu can run each thread > in pset on processor removal and if not reset affinity of that thread.
I'd say that the removal could be simply refused if there is still some thread with an affinity mask that would be entirely disabled. And conversely, I'd say thread_set_affinity should refuse an affinity mask that would end up being entirely disabled. Samuel
