From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Chen, Kenneth W" Date: Tue, 24 Jan 2006 22:31:00 +0000 Subject: RE: [PATCH] SN2 user-MMIO CPU migration Message-Id: <200601242231.k0OMV0g15339@unix-os.sc.intel.com> List-Id: References: <20060118163305.Y42462@chenjesu.americas.sgi.com> In-Reply-To: <20060118163305.Y42462@chenjesu.americas.sgi.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-ia64@vger.kernel.org Brent Casavant wrote on Tuesday, January 24, 2006 1:13 PM > The new patch below implements pretty much exactly that. I'm not > thrilled about carrying around the thread_info.last_cpu field even > on non-SN2, but making the presence/updating of the field conditional > led to huge headaches in the "should we call platform_switch_from" > conditionals/logic. Chen, Kenneth wrote on Tuesday, January 24, 2006 1:44 PM > Yeah, I'm not thrilled either. Currently, on context switch kernel > doesn't read thread_info.cpu, nor any other neighboring field. With > the patch, it needs to do a cacheline read and possibly an update. > If unlucky, it might be two cache lines with last_cpu 92 bytes away > from thread_info.cpu. Perhaps, last_cpu should be next to cpu field. > On the other hand, it may not be too bad because on kernel exit, flags > field in thread_info is accessed for TIF_* and if kernel exit path is > taken back-to-back with context switch, the net effect maybe small. Thinking about it, I wonder if Ingo would entertain the idea of moving p->thread_info.cpu into p->thread.cpu. Ingo? - Ken