Re: [PATCH] x86/UV: GRU distributed mode mappings

[Ingo Molnar <mingo@kernel.org>]

* Dimitri Sivanich <sivanich@sgi.com> wrote:

> On Tue, May 28, 2013 at 09:59:12AM +0200, Ingo Molnar wrote:
> > 
> > * Dimitri Sivanich <sivanich@sgi.com> wrote:
> > 
> > > Setup mappings for GRU distributed mode and include them as untracked
> > > pat ranges.
> > 
> > Would be nice to have the customary changelog content:
> > 
> >   " Current code does (A), this has a problem when (B).
> >     We can improve this doing (C), because (D)."
> > 
> > Or so - your changelog is pretty light on the 'why' aspect.
> >
> How about the following?

Looks good to me.

> UV GRU hardware will support an optional distributed mode that will allow 
> per-node address mapping of local GRU space, as opposed to mapping all GRU
> hardware to the same contiguous high space.
> 
> If GRU distributed mode is selected, setup per-node page table mappings and
> include them as untracked pat ranges.
> 
> Signed-off-by: Dimitri Sivanich <sivanich@sgi.com>
> ---
>  arch/x86/kernel/apic/x2apic_uv_x.c |   63 +++++++++++++++++++++++++++++----
>  1 file changed, 57 insertions(+), 6 deletions(-)

So this looks like suitable v3.11 material - not an urgent fix needed for 
v3.10, right?

Noticed this detail:

> +	} else
> +		return start >= gru_start_paddr && end <= gru_end_paddr;

in the kernel we generally try to balance curly braces - either all 
branches are curly or none.

> +	/* Save upper (63:M) bits of address only for is_GRU_range */
> +	gru_first_node_paddr &= gru_dist_umask;
> +	gru_last_node_paddr &= gru_dist_umask;
> +	pr_info("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
> +		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);

I suspect this should be pr_debug()?

Thanks,

	Ingo