Date: Tue, 16 Mar 2010 12:38:20 -0400 From: Jung-uk Kim <jkim@FreeBSD.org> To: freebsd-hackers@freebsd.org Cc: Kostik Belousov <kostikbel@gmail.com>, Marcel Moolenaar <xcllnt@mac.com> Subject: Re: [RFC] DTrace SYSCALL provider (was Re: [RFC] Saving the latest errno from syscalls.) Message-ID: <201003161238.23768.jkim@FreeBSD.org> In-Reply-To: <201003151410.35959.jkim@FreeBSD.org> References: <201003111624.51018.jkim@FreeBSD.org> <201003121332.16979.jkim@FreeBSD.org> <201003151410.35959.jkim@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Monday 15 March 2010 02:10 pm, Jung-uk Kim wrote: > On Friday 12 March 2010 01:32 pm, Jung-uk Kim wrote: > > On Friday 12 March 2010 04:29 am, Kostik Belousov wrote: > > > On Thu, Mar 11, 2010 at 06:15:07PM -0500, Jung-uk Kim wrote: > > > > On Thursday 11 March 2010 04:55 pm, Marcel Moolenaar wrote: > > > > > On Mar 11, 2010, at 1:24 PM, Jung-uk Kim wrote: > > > > > > While I was debugging syscalls, I found a very useful > > > > > > field in struct thread, td_errno. It seems it was added > > > > > > for dtrace but it is only populated on amd64 and i386. > > > > > > Is the attached patch acceptable for maintainers of other > > > > > > platforms? > > > > > > > > > > Isn't it better to do it in cpu_set_syscall_retval()? > > > > > That way you catch all cases, plus you can save the > > > > > translated error as well... > > > > > > > > I just took amd64/i386 as an example and I was not sure > > > > whether it was meant to store translated error or not. Does > > > > anyone with DTrace internal knowledge answer the question? > > > > > > I do not know that much about DTrace, but it seems that setting > > > td_errno in cpu_set_syscall_retval() is too late. Dtrace has a > > > probe after the syscall return, and it is called right before > > > cpu_set_syscall_retval() can be reasonably called. The probe > > > only issued for syscall that goes into sysent. > > > > Ah, I can see that now. So, if/when we implement DTrace SYSCALL > > provider for other arches, this is the right place. :-) > > I went ahead and implemented DTrace SYSCALL providers for non-x86 > arches. It passes 'make universe' test but I don't know if it > works. Can maintainers of other arches test or review the attached > patch? I realized there are too many missing pieces for DTrace to work on other arches. :-( Please ignore this patch. Jung-uk Kim
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201003161238.23768.jkim>