Date: Tue, 28 Feb 2023 13:48:14 -0600 From: Mike Karels <mike@karels.net> To: John Baldwin <jhb@FreeBSD.org> Cc: Konstantin Belousov <kostikbel@gmail.com>, Mateusz Guzik <mjguzik@gmail.com>, markj@freebsd.org, src-committers@freebsd.org, dev-commits-src-all@freebsd.org, dev-commits-src-main@freebsd.org Subject: Re: git: 68912701700c - main - ffs_suspend.c: clean up includes Message-ID: <C4CB519A-FD0C-4B7E-A013-C37B89AA50D2@karels.net> In-Reply-To: <8ceb993e-1589-fd24-1786-4438acb9a07f@FreeBSD.org> References: <202212292056.2BTKuOIu047460@gitrepo.freebsd.org> <CAGudoHEqJVvxp8-PbbrHM9=Qz=jSV8PQR8q848HJOmULiVSZzg@mail.gmail.com> <Y7QXN18gBTKDMj8j@kib.kiev.ua> <8ceb993e-1589-fd24-1786-4438acb9a07f@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 28 Feb 2023, at 12:55, John Baldwin wrote: > On 1/3/23 3:53 AM, Konstantin Belousov wrote: >> On Mon, Jan 02, 2023 at 06:29:57PM +0100, Mateusz Guzik wrote: >>> On 12/29/22, Konstantin Belousov <kib@freebsd.org> wrote: >>>> The branch main has been updated by kib: >>>> >>>> URL: >>>> https://cgit.FreeBSD.org/src/commit/?id=3D68912701700ca3230f3e2d4b78= 58a038f884a327 >>>> >>>> commit 68912701700ca3230f3e2d4b7858a038f884a327 >>>> Author: Konstantin Belousov <kib@FreeBSD.org> >>>> AuthorDate: 2022-12-28 18:17:53 +0000 >>>> Commit: Konstantin Belousov <kib@FreeBSD.org> >>>> CommitDate: 2022-12-29 20:55:39 +0000 >>>> >>>> ffs_suspend.c: clean up includes >>>> >>>> Order includes alphabetically. >>>> Remove unneeded sys/param.h, it is already included by sys/syst= m.h. >>>> >>>> Reviewed by: mckusick >>>> Sponsored by: The FreeBSD Foundation >>>> MFC after: 1 week >>>> Differential revision: https://reviews.freebsd.org/D37896 >>>> --- >>>> sys/ufs/ffs/ffs_suspend.c | 7 +++---- >>>> 1 file changed, 3 insertions(+), 4 deletions(-) >>>> >>>> diff --git a/sys/ufs/ffs/ffs_suspend.c b/sys/ufs/ffs/ffs_suspend.c >>>> index d13097109758..e7c976b6e921 100644 >>>> --- a/sys/ufs/ffs/ffs_suspend.c >>>> +++ b/sys/ufs/ffs/ffs_suspend.c >>>> @@ -33,15 +33,14 @@ >>>> #include <sys/cdefs.h> >>>> __FBSDID("$FreeBSD$"); >>>> >>>> -#include <sys/param.h> >>>> #include <sys/systm.h> >>>> #include <sys/buf.h> >>>> -#include <sys/ioccom.h> >>>> -#include <sys/mount.h> >>>> -#include <sys/vnode.h> >>>> #include <sys/conf.h> >>>> +#include <sys/ioccom.h> >>>> #include <sys/jail.h> >>>> +#include <sys/mount.h> >>>> #include <sys/sx.h> >>>> +#include <sys/vnode.h> >>>> >>>> #include <security/mac/mac_framework.h> >>>> >>>> >>> >>> tinderbox fails the KCSAN et al kernels: >>> >>> In file included from /usr/src/sys/ufs/ffs/ffs_suspend.c:36: >>> In file included from /usr/src/sys/sys/systm.h:44: >>> In file included from ./machine/atomic.h:73: >>> /usr/src/sys/sys/atomic_san.h:117:24: error: unknown type name 'uint8= _t' >>> ATOMIC_SAN_FUNCS(char, uint8_t); >>> ^ >>> it bisects to this commit >>> >> So the problem is that sys/systm.h includes machine/atomic.h which alw= ays >> had the pre-requisite sys/types.h, and this is documented in atomic(9)= =2E >> But sys/atomic_san.h uses C89 types in prototypes, not just macros, an= d >> this breaks for real. >> >> I can >> 1. Just add sys/types.h to ufs_suspend.c (I prefer not) >> 2. Add sys/types.h to sys/atomic_san.h. >> 3. Add sys/types.h to sys/systm.h. >> >> IMO #2 is not the best solution, since it pollutes systm.h only someti= mes, >> when the right kernel options are used. I would prefer #3, it seems, = but >> want to ensure that there is a consensus about the addition to sys/sys= tm.h. > > FYI, I got a different ream of breakages downstream in CheriBSD due to = this > commit removing param.h as well. I also only ever recalled Bruce telli= ng me > to do either <sys/types.h> or <sys/param.h> first (and only one of them= ). > I don't recall any special rules for <sys/systm.h>. > > CheriBSD has a local diff that tries to sort headers in systm.h which i= s why > we get larger breakage: > > --- a/sys/sys/systm.h > +++ b/sys/sys/systm.h > @@ -41,13 +41,22 @@ > #define _SYS_SYSTM_H_ > #include <sys/cdefs.h> > -#include <machine/atomic.h> > -#include <machine/cpufunc.h> > #include <sys/callout.h> > #include <sys/kassert.h> > #include <sys/queue.h> > #include <sys/stdint.h> /* for people using printf main= ly */ > +#include <machine/atomic.h> > +#include <machine/cpufunc.h> > > I think if we want <sys/systm.h> to be self-contained we should swap <s= ys/cdefs.h> > for <sys/types.h>. I think moving the <machine/*> includes down would = also be > more consistent with style(9) as well. I prefer retaining <sys/param.h> or <sys/types.h> to be the first include= in kernel files, as has long been traditional. systm.h didn=E2=80=99t originally i= nclude param.h, and param.h isn=E2=80=99t anywhere near the top of systm.h. I=E2=80=99m = sure it was just added to unbreak some file that didn=E2=80=99t include param.h when a dependenc= y on MAXCPU crept in. It looks like 15 or more headers in sys/ include <sys/param.h>; syst= m.h is not really special in that regard. Finally, style(9) documents use of ei= ther <sys/param.h> or <sys/types.h> first, then <sys/systm.h> if needed. Mike
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C4CB519A-FD0C-4B7E-A013-C37B89AA50D2>