Tonight I put on a Jon Hopkins album, in recollection of my UK ARM Linux colleagues perhaps, and started testing, and eventually wound up with this one:
commit 621a0147d5c921f4cc33636ccd0602ad5d7cbfbc Author: Will Deacon <[email protected]> Date: Wed Jun 12 12:25:56 2013 +0100 ARM: 7757/1: mm: don't flush icache in switch_mm with hardware broadcasting When scheduling an mm on a CPU where it hasn't previously been used, we flush the icache on that CPU so that any code loaded previously on a different core can be safely executed. For cores with hardware broadcasting of cache maintenance operations, this is clearly unnecessary, since the inner-shareable invalidation in __sync_icache_dcache will affect all CPUs. This patch conditionalises the icache flush in switch_mm based on cache_ops_need_broadcast(). Acked-by: Catalin Marinas <[email protected]> Reported-by: Albin Tonnerre <[email protected]> Signed-off-by: Will Deacon <[email protected]> Signed-off-by: Russell King <[email protected]> ... v3.11-rc2 boots with it reverted. What also works is v3.11-rc2 with the below patch applied. Would be pleased to boot-test anything you'd care to send my way, as long as you can tolerate response latency jitter. - Paul diff --git a/arch/arm/include/asm/mmu_context.h b/arch/arm/include/asm/mmu_context.h index b5792b7..8dfb295 100644 --- a/arch/arm/include/asm/mmu_context.h +++ b/arch/arm/include/asm/mmu_context.h @@ -112,7 +112,7 @@ switch_mm(struct mm_struct *prev, struct mm_struct *next, * so check for possible thread migration and invalidate the I-cache * if we're new to this CPU. */ - if (cache_ops_need_broadcast() && + if (1 && !cpumask_empty(mm_cpumask(next)) && !cpumask_test_cpu(cpu, mm_cpumask(next))) __flush_icache_all(); -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
