[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Bringing up sequence for non-boot CPU fails

On Tue, 2014-02-18 at 18:33 +0000, Julien Grall wrote:
> On 02/18/2014 06:17 PM, Oleksandr Tyshchenko wrote:
> > Ian,
> Hello Oleksandr,
> > I have checked your suggestion with full cache flush.
> > For this purposes I have used ARMV7 specific function from our U-Boot.
> > This function performs clean and invalidation of the entire data cache
> > at all levels.
> Did you try to only clean the cache? When page table for the secondary
> CPU is created Xen only clean the cache for the specific range.
> I suspect it's not enough and we need to invalidate.

I think the error is occurring when paging is first enabled with HTTBR
set to boot_pgtable, which the secondary CPU is setting up itself during
head.S. The switch to the secondaries own PTs, which are setup by the
boot processor for it, happens a bit later. Other than the boot_pgtable
setup that the CPUs do themselves I don't think anything is touching the
boot pts with caches turned on.

So long as the data is not modified between the clean and enabling
MMU/caches the invalidate shouldn't really matter -- any data which
magically appears when you enable the caches should be in sync with the
underlying RAM I think.

BTW, it would be worth just confirming that caches are not enabled
either upon entry to Xen or after the magic smc #0.

> It should be easy to try with this small patch:
> diff --git a/xen/include/asm-arm/page.h b/xen/include/asm-arm/page.h
> index e00be9e..5a8aba2 100644
> --- a/xen/include/asm-arm/page.h
> +++ b/xen/include/asm-arm/page.h
> @@ -234,7 +234,7 @@ static inline void clean_xen_dcache_va_range(void
> *p, unsigned long size)
>      void *end;
>      dsb();           /* So the CPU issues all writes to the range */
>      for ( end = p + size; p < end; p += cacheline_bytes )
> -        asm volatile (__clean_xen_dcache_one(0) : : "r" (p));
> +        asm volatile (__clean_and_invalidate_xen_dcache_one(0) : : "r"
> (p));
>      dsb();           /* So we know the flushes happen before continuing */
>  }
> Regards,

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.