From: linux@arm.linux.org.uk (Russell King - ARM Linux) Date: Sun, 14 Feb 2010 00:18:58 +0000 Subject: [PATCH] Help fix SMP boot issues with L2 cache In-Reply-To: <4B7739F7.7060604@arm.com> References: <20100212182552.GF18595@n2100.arm.linux.org.uk> <4B7739F7.7060604@arm.com> Message-ID: <20100214001858.GA16553@n2100.arm.linux.org.uk> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Sat, Feb 13, 2010 at 11:47:03PM +0000, Catalin Marinas wrote: > Russell King - ARM Linux wrote: >> ARM: Add L2 cache handling to smp boot support >> The page table and secondary data which we're asking the >> secondary CPU >> to make use of has to hit RAM to ensure that the secondary CPU can see >> it since it may not be taking part in coherency or cache searches at >> this point. >> Signed-off-by: Russell King >> >> arch/arm/kernel/smp.c | 4 +++- >> 1 files changed, 3 insertions(+), 1 deletions(-) >> >> diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c >> index 57162af..577543f 100644 >> --- a/arch/arm/kernel/smp.c >> +++ b/arch/arm/kernel/smp.c >> @@ -99,6 +99,7 @@ int __cpuinit __cpu_up(unsigned int cpu) >> *pmd = __pmd((PHYS_OFFSET & PGDIR_MASK) | >> PMD_TYPE_SECT | PMD_SECT_AP_WRITE); >> flush_pmd_entry(pmd); >> + outer_clean_range(__pa(pmd), __pa(pmd + 1)); >> /* >> * We need to tell the secondary core where to find >> @@ -106,7 +107,8 @@ int __cpuinit __cpu_up(unsigned int cpu) >> */ >> secondary_data.stack = task_stack_page(idle) + THREAD_START_SP; >> secondary_data.pgdir = virt_to_phys(pgd); >> - wmb(); >> + __cpuc_flush_dcache_area(&secondary_data, sizeof(secondary_data)); >> + outer_clean_range(__pa(&secondary_data), __pa(&secondary_data + 1)); >> /* >> * Now bring the CPU into our world. > > But is the outer_cache initialised at this point? If it isn't, the calls are no-ops, so are harmless. However, there are platforms where the L2 is already initialized at this point, and therefore this needs to be handled. Whether platforms have initialized their L2 cache support while the hardware is enabled is their own lookout at this point - that's outside of the scope of this patch. > I recall for RealView > I had to change the l2x0_init to an early_initcall: > > http://linux-arm.org/git?p=linux-2.6-stable.git;a=commitdiff;h=70e1b8d308f635eaeb7dad14fc5c79f082a147ef;hp=f034a4ab31b0663a5b68c65503122bed1eaca349 And that's what I'm addressing with this patch - the need to patch every platsmp.c file with that kind of dirty hack. With my patch, the SMP generic data structures will be covered by the appropriate code, rather than having every SMP platform having to do that itself. Please update your patch for mine - the only thing you need to look after is 'pen_release'.