From 4d803fcdcd97dd346d4b39c3b76e5879cead8a31 Mon Sep 17 00:00:00 2001 From: "David S. Miller" Date: Thu, 8 Sep 2005 14:37:53 -0700 Subject: [SPARC64]: Inline membar()'s again. Since GCC has to emit a call and a delay slot to the out-of-line "membar" routines in arch/sparc64/lib/mb.S it is much better to just do the necessary predicted branch inline instead as: ba,pt %xcc, 1f membar #whatever 1: instead of the current: call membar_foo dslot because this way GCC is not required to allocate a stack frame if the function can be a leaf function. This also makes this bug fix easier to backport to 2.4.x Signed-off-by: David S. Miller --- include/asm-sparc64/system.h | 49 +++++++++++++++++++++++++++++++++++++------- 1 file changed, 42 insertions(+), 7 deletions(-) (limited to 'include') diff --git a/include/asm-sparc64/system.h b/include/asm-sparc64/system.h index 5e94c05dc2fc..b5417529f6f1 100644 --- a/include/asm-sparc64/system.h +++ b/include/asm-sparc64/system.h @@ -28,13 +28,48 @@ enum sparc_cpu { #define ARCH_SUN4C_SUN4 0 #define ARCH_SUN4 0 -extern void mb(void); -extern void rmb(void); -extern void wmb(void); -extern void membar_storeload(void); -extern void membar_storeload_storestore(void); -extern void membar_storeload_loadload(void); -extern void membar_storestore_loadstore(void); +/* These are here in an effort to more fully work around Spitfire Errata + * #51. Essentially, if a memory barrier occurs soon after a mispredicted + * branch, the chip can stop executing instructions until a trap occurs. + * Therefore, if interrupts are disabled, the chip can hang forever. + * + * It used to be believed that the memory barrier had to be right in the + * delay slot, but a case has been traced recently wherein the memory barrier + * was one instruction after the branch delay slot and the chip still hung. + * The offending sequence was the following in sym_wakeup_done() of the + * sym53c8xx_2 driver: + * + * call sym_ccb_from_dsa, 0 + * movge %icc, 0, %l0 + * brz,pn %o0, .LL1303 + * mov %o0, %l2 + * membar #LoadLoad + * + * The branch has to be mispredicted for the bug to occur. Therefore, we put + * the memory barrier explicitly into a "branch always, predicted taken" + * delay slot to avoid the problem case. + */ +#define membar_safe(type) \ +do { __asm__ __volatile__("ba,pt %%xcc, 1f\n\t" \ + " membar " type "\n" \ + "1:\n" \ + : : : "memory"); \ +} while (0) + +#define mb() \ + membar_safe("#LoadLoad | #LoadStore | #StoreStore | #StoreLoad") +#define rmb() \ + membar_safe("#LoadLoad") +#define wmb() \ + membar_safe("#StoreStore") +#define membar_storeload() \ + membar_safe("#StoreLoad") +#define membar_storeload_storestore() \ + membar_safe("#StoreLoad | #StoreStore") +#define membar_storeload_loadload() \ + membar_safe("#StoreLoad | #LoadLoad") +#define membar_storestore_loadstore() \ + membar_safe("#StoreStore | #LoadStore") #endif -- cgit v1.2.2