From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 27497 invoked by alias); 18 Feb 2011 20:24:46 -0000 Mailing-List: contact archer-help@sourceware.org; run by ezmlm Sender: Precedence: bulk List-Post: List-Help: List-Subscribe: List-Id: Received: (qmail 27485 invoked by uid 22791); 18 Feb 2011 20:24:45 -0000 X-SWARE-Spam-Status: No, hits=-6.1 required=5.0 tests=AWL,BAYES_00,RCVD_IN_DNSWL_HI,SARE_SUB_IMPROVE,SPF_HELO_PASS,T_RP_MATCHES_RCVD X-Spam-Check-By: sourceware.org Date: Fri, 18 Feb 2011 20:24:00 -0000 From: Oleg Nesterov To: Roland McGrath Cc: Jan Kratochvil , Project Archer Subject: Re: ptrace improvement ideas (QPassSignals) Message-ID: <20110218201630.GA12693@redhat.com> References: <20110203223905.D0C77180081@magilla.sf.frob.com> <20110216113110.GA18421@host1.dyn.jankratochvil.net> <20110216183608.4A8E01806E0@magilla.sf.frob.com> <20110216201306.GC17601@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20110216201306.GC17601@redhat.com> User-Agent: Mutt/1.5.18 (2008-05-17) X-SW-Source: 2011-q1/txt/msg00079.txt.bz2 On 02/16, Oleg Nesterov wrote: > > On 02/16, Roland McGrath wrote: > > > It's already desireable to restructure > > the ptrace implementation details so that ptrace-related storage is not > > allocated when ptrace is not being used, > > Yes. IIRC, we already had the patches, but right now I can't recall > if they were complete (most probably yes, except I am not sure about > task->ptrace member). I'll try to find this series. It turns out, it was complete but it should be changed a bit. > Probably this is the first thing we should do. Yes, I think this can help to implement the new features. But there is a problem. If we want PTRACE_ATTCH_ALL_THREADS (and iiuc we do want) we have to use GFP_ATOMIC to allocate ptrace_state under tasklist_lock. I wonder if we can invent a sleepable lock. Actually it is also needed for cgroups, the guys from google are going to add the rw_semaphore into signal_struct. But it depends on CONFIG_CGROUPS and doesn't cover exit. Yes, yes, I am talking about implementation details again... Oleg.