[PATCH v2 2/8] arm/tegra: prepare early init for multiple tegra variants
Peter De Schrijver
pdeschrijver at nvidia.com
Mon Nov 7 07:36:30 EST 2011
On Thu, Nov 03, 2011 at 10:21:16AM +0100, Peter De Schrijver wrote:
> On Wed, Nov 02, 2011 at 07:30:43PM +0100, Colin Cross wrote:
> > On Wed, Nov 2, 2011 at 7:09 AM, Peter De Schrijver
> > <pdeschrijver at nvidia.com> wrote:
> > > This patch splits the early init code in a common and a tegra20 specific part.
> > > L2 cache initialization is generalized and discovers the cache associativity
> > > at runtime.
> > >
> > > Signed-off-by: Peter De Schrijver <pdeschrijver at nvidia.com>
> > > ---
> > <snip>
> >
> > > -static __initdata struct tegra_clk_init_table common_clk_init_table[] = {
> > > +void (*arch_reset)(char mode, const char *cmd) = tegra_assert_system_reset;
> > > +
> > This is unrelated to the patch description?
> >
>
> This is a left over from the past. Will be removed from the next version.
>
> > > +static __initdata struct tegra_clk_init_table tegra20_clk_init_table[] = {
> > > /* name parent rate enabled */
> > > { "clk_m", NULL, 0, true },
> > > { "pll_p", "clk_m", 216000000, true },
> > > @@ -59,25 +61,34 @@ static __initdata struct tegra_clk_init_table common_clk_init_table[] = {
> > > { NULL, NULL, 0, 0},
> > > };
> > >
> > > -static void __init tegra_init_cache(void)
> > > +static void __init tegra_init_cache(u32 tag_latency, u32 data_latency)
> > > {
> > > #ifdef CONFIG_CACHE_L2X0
> > > void __iomem *p = IO_ADDRESS(TEGRA_ARM_PERIF_BASE) + 0x3000;
> > > + u32 aux_ctrl;
> > > +
> > > + writel_relaxed(tag_latency, p + L2X0_TAG_LATENCY_CTRL);
> > > + writel_relaxed(data_latency, p + L2X0_DATA_LATENCY_CTRL);
> > >
> > > - writel_relaxed(0x331, p + L2X0_TAG_LATENCY_CTRL);
> > > - writel_relaxed(0x441, p + L2X0_DATA_LATENCY_CTRL);
> > > + aux_ctrl = readl(p + L2X0_CACHE_TYPE);
> > > + aux_ctrl = (aux_ctrl & 0x700) << (17-8);
> >
> > If I'm interpreting the PL310 TRM correctly, the cache associativity
> > value in the L2X0_CACHE_TYPE register is read from the aux control
> > register, so this is just leaving the existing value in place. If the
> > value is not being modified, just take it out of the aux_ctrl mask
> > passed to l2x0_init.
> >
>
> This seems to work fine on our android kernels though. Need to dig deeper
> what's going on here.
Doesn't seem to be that simple on tegra. The L2X0_CACHE_TYPE can indicate a
different waysize depending on fuses.
Cheers,
Peter.
More information about the linux-arm-kernel
mailing list