From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Message-ID: <17506.31456.68099.57515@cargo.ozlabs.ibm.com> Date: Thu, 11 May 2006 09:44:32 +1000 From: Paul Mackerras To: Richard Henderson , t@twiddle.net, linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org, linuxppc-dev@ozlabs.org, amodra@bigpond.net Subject: Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables In-Reply-To: <17506.29128.591758.502430@cargo.ozlabs.ibm.com> References: <17505.26159.807484.477212@cargo.ozlabs.ibm.com> <20060510154702.GA28938@twiddle.net> <17506.29128.591758.502430@cargo.ozlabs.ibm.com> List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , I wrote: > Hmmm... Would it be sufficient to use a RELOC_HIDE in __get_cpu_var, > like this? > > #define __get_cpu_var(x) (*(RELOC_HIDE(&per_cpu__##x, 0))) But that won't work because the compiler can still cache &per_cpu__x. I guess I have to do this: #define __get_cpu_var(var, cpu) \ (*(__typeof__(&per_cpu__##var))({ \ void *__ptr; \ asm("addi %0,13,per_cpu__"#var"@tprel" \ : "=r" (__ptr)); \ __ptr; \ })) That means we lose the possible optimization of combining the addi into a following load or store. Bah. However, I guess it's still better than what we do at the moment. Paul.