From mboxrd@z Thu Jan 1 00:00:00 1970 X-Msuck: nntp://news.gmane.org/gmane.linux.lib.musl.general/9432 Path: news.gmane.org!not-for-mail From: Rich Felker Newsgroups: gmane.linux.lib.musl.general Subject: Re: [PATCH] add sched_getcpu, with vDSO support Date: Wed, 2 Mar 2016 00:55:42 -0500 Message-ID: <20160302055542.GN9349@brightrain.aerifal.cx> References: <1456839940-10330-1-git-send-email-nathan@nathan7.eu> <1456847766-23953-1-git-send-email-nathan@nathan7.eu> Reply-To: musl@lists.openwall.com NNTP-Posting-Host: plane.gmane.org Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Trace: ger.gmane.org 1456898160 30524 80.91.229.3 (2 Mar 2016 05:56:00 GMT) X-Complaints-To: usenet@ger.gmane.org NNTP-Posting-Date: Wed, 2 Mar 2016 05:56:00 +0000 (UTC) To: musl@lists.openwall.com Original-X-From: musl-return-9445-gllmg-musl=m.gmane.org@lists.openwall.com Wed Mar 02 06:55:56 2016 Return-path: Envelope-to: gllmg-musl@m.gmane.org Original-Received: from mother.openwall.net ([195.42.179.200]) by plane.gmane.org with smtp (Exim 4.69) (envelope-from ) id 1aazlM-0003Vx-LO for gllmg-musl@m.gmane.org; Wed, 02 Mar 2016 06:55:56 +0100 Original-Received: (qmail 3579 invoked by uid 550); 2 Mar 2016 05:55:54 -0000 Mailing-List: contact musl-help@lists.openwall.com; run by ezmlm Precedence: bulk List-Post: List-Help: List-Unsubscribe: List-Subscribe: List-ID: Original-Received: (qmail 3555 invoked from network); 2 Mar 2016 05:55:54 -0000 Content-Disposition: inline In-Reply-To: <1456847766-23953-1-git-send-email-nathan@nathan7.eu> User-Agent: Mutt/1.5.21 (2010-09-15) Original-Sender: Rich Felker Xref: news.gmane.org gmane.linux.lib.musl.general:9432 Archived-At: On Tue, Mar 01, 2016 at 04:56:06PM +0100, Nathan Zadoks wrote: > This is a GNU extension, but a fairly minor one, for a system call that > otherwise has no libc wrapper. > > Adding it was discussed previously, without any objections: > http://www.openwall.com/lists/musl/2015/05/08/24 > --- > > Whoops, forgot to make getcpu_init a static function in the previous version. > Fixed! > > --- > > arch/x86_64/syscall_arch.h | 2 ++ > include/sched.h | 1 + > src/sched/sched_getcpu.c | 45 +++++++++++++++++++++++++++++++++++++++++++++ > 3 files changed, 48 insertions(+) > create mode 100644 src/sched/sched_getcpu.c > > diff --git a/arch/x86_64/syscall_arch.h b/arch/x86_64/syscall_arch.h > index a7a7b5a..54e05ff 100644 > --- a/arch/x86_64/syscall_arch.h > +++ b/arch/x86_64/syscall_arch.h > @@ -64,3 +64,5 @@ static __inline long __syscall6(long n, long a1, long a2, long a3, long a4, long > #define VDSO_USEFUL > #define VDSO_CGT_SYM "__vdso_clock_gettime" > #define VDSO_CGT_VER "LINUX_2.6" > +#define VDSO_GETCPU_SYM "__vdso_getcpu" > +#define VDSO_GETCPU_VER "LINUX_2.6" > diff --git a/include/sched.h b/include/sched.h > index 3e34a72..7e88f09 100644 > --- a/include/sched.h > +++ b/include/sched.h > @@ -76,6 +76,7 @@ void free(void *); > > typedef struct cpu_set_t { unsigned long __bits[128/sizeof(long)]; } cpu_set_t; > int __sched_cpucount(size_t, const cpu_set_t *); > +int sched_getcpu(void); > int sched_getaffinity(pid_t, size_t, cpu_set_t *); > int sched_setaffinity(pid_t, size_t, const cpu_set_t *); > > diff --git a/src/sched/sched_getcpu.c b/src/sched/sched_getcpu.c > new file mode 100644 > index 0000000..096f06f > --- /dev/null > +++ b/src/sched/sched_getcpu.c > @@ -0,0 +1,45 @@ > +#define _GNU_SOURCE > +#include > +#include > +#include > +#include "syscall.h" > +#include "atomic.h" > + > +#ifdef VDSO_GETCPU_SYM > + > +void *__vdsosym(const char *, const char *); > + > +static void *volatile vdso_func; > + > +typedef long (*getcpu_f)(unsigned *, unsigned *, void *); > + > +static long getcpu_init(unsigned *cpu, unsigned *node, void *unused) > +{ > + void *p = __vdsosym(VDSO_GETCPU_VER, VDSO_GETCPU_SYM); > + getcpu_f f = (getcpu_f)p; > + a_cas_p(&vdso_func, (void *)getcpu_init, p); > + return f ? f(cpu, node, unused) : -ENOSYS; > +} > + > +static void *volatile vdso_func = (void *)getcpu_init; > + > +#endif > + > +int sched_getcpu(void) > +{ > + int r; > + unsigned cpu; > + > +#ifdef VDSO_CGT_SYM Wrong macro in the #ifdef. > + getcpu_f f = (getcpu_f)vdso_func; > + if (f) { > + r = f(&cpu, NULL, NULL); Not a big deal, but usually in musl we use 0 rather than the NULL macro. Actually I wondered if the function actually needs to take the useless extra 2 arguments, but I think for the sake of correctness it's best to do it this way -- the callee in the vdso has 3 args, so it should be called with the correct type. > + if (!r) return cpu; > + if (r != -ENOSYS) return __syscall_ret(r); > + } > +#endif > + > + r = __syscall(SYS_getcpu, &cpu, NULL, NULL); > + if (!r) return cpu; > + return __syscall_ret(r); > +} Also tabs should be used for indention, not spaces. One other thing I thought might be nice is initially committing the trivial syscall-only version that adds the public prototype, then doing vdso support as a separate commit, but if that's a pain to do don't worry about it. Rich