From mboxrd@z Thu Jan 1 00:00:00 1970 From: Russ Anderson Date: Thu, 16 Sep 2004 18:33:13 +0000 Subject: Re: [Patch] Per CPU MCA/INIT data save areas Message-Id: <200409161833.i8GIXDaS030189@ben.americas.sgi.com> List-Id: References: <200408312112.i7VLCPc9002392@ben.americas.sgi.com> In-Reply-To: <200408312112.i7VLCPc9002392@ben.americas.sgi.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-ia64@vger.kernel.org Keith Owens wrote: > On Wed, 15 Sep 2004 09:28:58 -0700, > "Luck, Tony" wrote: > >>Unless anyone has a better solution, I think Keith's suggestion is the > >>next best alternative. It is no worse than the current code > >>(in terms of off-node access) and allows forward progress on the rest > >>of the code. > > > >Take a look at the MCA-TLB recovery code. It saves per-cpu > >information in the "ia64_mca_tlb_list[]" array, and searches > >through it for an entry that matches cr.lid at fault time. > > > >You could add extra element(s) to this structure for the percpu > >areas. > > What about storing the physical address of the per-cpu data area in > ar.k3? Given that the MCA/INIT paths are (by definition) not hot paths, and that CPU regs are a scarce resource, I think there are better places to burn a CPU reg on. Adding a pointer (or two) to the ia64_mca_tlb_info structure would give the necessary functionality. -- Russ Anderson, OS RAS/Partitioning Project Lead SGI - Silicon Graphics Inc rja@sgi.com