[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <nycvar.YFH.7.76.2001101530050.31058@cbobk.fhfr.pm>
Date: Fri, 10 Jan 2020 15:31:10 +0100 (CET)
From: Jiri Kosina <jikos@...nel.org>
To: Linus Torvalds <torvalds@...ux-foundation.org>
cc: Marcel Holtmann <marcel@...tmann.org>,
Dmitry Torokhov <dmitry.torokhov@...il.com>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Benjamin Tissoires <benjamin.tissoires@...hat.com>
Subject: Re: [GIT PULL] HID fixes
On Thu, 9 Jan 2020, Linus Torvalds wrote:
> > Now, good source code presumably notices EPOLLERR and handles it. So
> > it _shouldn't_ matter what the kernel does if an error occurs. I
> > haven't checked what people _actually_ do, tnough. I worry sometimes
> > that user space just looks at EPOLLIN sees it not being set, and gets
> > stuck in a busy loop polling in case of errors.
>
> Googling around for it, I find this, for example:
>
> https://github.com/scylladb/seastar/issues/309
>
> and yes, I think that's technically a user space bug, but it's very
> much an example of this: they expect to get errors through read() or
> write() calls, and get confused when poll() does not say that the fd
> is readable or writable.
>
> I don't know how common this is, but it didn't take a _lot_ of
> googling for me to find that one..
Right, I think it's quite a convicing argument, and the issue is rather
easy to avoid. I'll fix that up in the patch and send a fixup pull request
to you later today.
Marcel, please speak up if you have other plans.
--
Jiri Kosina
SUSE Labs
Powered by blists - more mailing lists