From: David Miller Date: Thu, 09 Apr 2009 05:37:19 +0000 Subject: [PATCH 0/12]: Sparc64 dynamic per-cpu support. Message-Id: <20090408.223719.61587584.davem@davemloft.net> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: tj@kernel.org Cc: sparclinux@vger.kernel.org, linux-kernel@vger.kernel.org This patch set converts sparc64 over to use the dynamic per-cpu stuff. Most of the work is the boring task of getting rid of the special "real_setup_per_cpu_areas()" sparc64 uses to setup the per-cpu areas before bootmem is available. We needed to do this because we initialized the cpu possible map at the same time as the stuff stored in the per-cpu cpu_data array. So we split those tasks up. Otherwise we wouldn't be able to convert to dynamic per-cpu because that stuff requires that bootmem and the page structs are ready to use. For now I'm using the draconian unoptimized x86 NUMA remap scheme. I really want to do this better because it is insanely suboptimal on Niagara chips where in some cases 8 cpus share the same physical TLB. This is against Linus's current tree and it's also in sparc-next-2.6