Messages in this thread | | | Date | Wed, 20 Jun 2018 04:33:21 +0100 | From | Al Viro <> | Subject | Re: [PATCH] __poll_t regressions this cycle |
| |
On Wed, Jun 20, 2018 at 10:57:41AM +0900, Linus Torvalds wrote: > On Mon, Jun 18, 2018 at 10:47 PM Al Viro <viro@zeniv.linux.org.uk> wrote: > > > > Do you prefer it in one commit, or split into per-driver pieces? > > If it's all the exact same issue with the same commit message, then > just one commit.
Some EPOLL... misspelled as POLL..., some ->poll() instances declared as returning unsigned int instead of __poll_t and some __poll_t local variables declared unsigned int. The only case beyond misannotations is POLLWRNORM instead of EPOLLWRNORM - basically, on sparc, mips et.al. EPOLLWRNORM (via epoll(7)) doesn't trigger for mei and xsk. Which used to be the case for all drivers (only EPOLLOUT used to trigger on these architectures) before the fixes this February...
| |