Re: [Xen-devel] [PATCH 3/4] xen: arm: add barrier before sev in smp_spin_table_cpu_up

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Mon, 2014-03-17 at 14:18 +0000, Julien Grall wrote:
> Hi Ian,
> 
> On 03/17/2014 01:52 PM, Ian Campbell wrote:
> > This ensures that the writeq to the release address has occurred.
> > 
> > In reality there is a dsb() in the iounmap() (in the eventual write_pte()) 
> > but
> > make it explicit.
> 
> There other place (see xen/arch/arm/platforms/exynos5.c) where we rely
> on the dsb in write_pte. I think we should move the dsb in iounmap().

Now you put it like that it does seem logical that any writes would be
issued prior to/as part of the iounmap.

In that context I think it would in fact be fine for iounmap to rely on
the dsb in write_pte since both a "change the mappings" functions. IOW
maybe we don't need to do anything after all.

> > The ARMv8 ARM recommends that sev() is usually accompanied by a dsb(), the
> > only other uses are in the v7 spinlock code which includes a dsb() already.
> 
> Would it be better to add dsb in the sev() macro?

ARM32 has a dsb_sev macro, maybe we should do the same for ARM64.

Ian


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel