From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: MIME-Version: 1.0 In-Reply-To: <5a084345-a8bf-3373-cc78-3d5d50acb13b@kernel.dk> References: <20170501184820.31025-1-tkusumi@tuxera.com> <20170501204421.GA13925@kernel.dk> <5a084345-a8bf-3373-cc78-3d5d50acb13b@kernel.dk> From: Tomohiro Kusumi Date: Tue, 2 May 2017 00:10:55 +0300 Message-ID: Subject: Re: [PATCH 01/11] configure: Disable CONFIG_POSIX_FALLOCATE on NetBSD even if posix_fallocate(3) compiles Content-Type: text/plain; charset=UTF-8 To: Jens Axboe Cc: fio@vger.kernel.org, Tomohiro Kusumi List-ID: 2017-05-01 23:57 GMT+03:00 Jens Axboe : > On 05/01/2017 02:54 PM, Tomohiro Kusumi wrote: >> Hi >> >> 2017-05-01 23:44 GMT+03:00 Jens Axboe : >>> On Mon, May 01 2017, kusumi.tomohiro@gmail.com wrote: >>>> From: Tomohiro Kusumi >>>> >>>> fio on NetBSD may have CONFIG_POSIX_FALLOCATE configuration enabled >>>> since posix_fallocate(3) compiles (at least on recent versions), >>>> but this is actually not supported on UFS as mentioned in below wiki >>>> and fio result. >>>> >>>> https://wiki.netbsd.org/projects/project/ffs-fallocate/ >>>>> This functionality is not currently implemented for FFS; >>>> >>>> compile_prog() during ./configure fails to catch this as it doesn't >>>> run the test code after compilation (and it needs to use a valid fd >>>> in order to do runtime test). >>>> >>>> This commit simply disables CONFIG_POSIX_FALLOCATE on NetBSD regardless >>>> of compilation result on ./configure. It doesn't check the fs type, >>>> but it should be enough provided that UFS is the fs used by majority >>>> of users and there's also no real alternative for disk fs. >>> >>> I'd much rather we just avoid it if we get EOPNOTSUPP at runtime. That >>> way we don't have to change anything later on, when/if NetBSD does get >>> support for fallocate. >> >> Yes, so the next one 2/11 actually mostly reverts this, and replace it >> with runtime check, so that it works not only with this specific >> NetBSD/UFS case, but also for e.g. Linux fs (something other than >> ext4, XFS, etc). >> I wasn't sure if runtime check (execute $TMPE) is preferred. > > Let's just drop 1/11 then, and have the first patch cover the runtime of > it. But don't do that from configure. Whatever fs is hosting configure > is less interesting. Maybe that doesn't support fallocate, but whatever > you end up running on does. Or vice versa. Thanks. Yes, you're right. They could certainly be different fs... > >> Any how, I'll resend the 1,2,3,4,11 with --cover-letter as you commented. > > Thanks! > > -- > Jens Axboe >