Messages in this thread | | | Subject | Re: [RFC v1 2/7] clk: meson: meson8b: do not use cpu_div3 for cpu_scale_out_sel | From | Jerome Brunet <> | Date | Thu, 15 Nov 2018 10:42:42 +0100 |
| |
On Wed, 2018-11-14 at 23:57 +0100, Martin Blumenstingl wrote: > The cpu_div3 clock (cpu_in divided by 3) generates a signal with a duty > cycle of 33%. The CPU clock however requires a clock signal with a duty > cycle of 50% to run stable. > cpu_div3 was observed to be problematic when cycling through all > available CPU frequencies (with additional patches on top of this one) > while running "stress --cpu 4" in the background. This caused sporadic > hangs where the whole system would fully lock up. > > Amlogic's 3.10 kernel code also does not use the cpu_div3 clock either > when changing the CPU clock. > > Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> > --- > drivers/clk/meson/meson8b.c | 11 +++++++++-- > 1 file changed, 9 insertions(+), 2 deletions(-) > > diff --git a/drivers/clk/meson/meson8b.c b/drivers/clk/meson/meson8b.c > index 40e77fe4ba7c..8a3c346e110d 100644 > --- a/drivers/clk/meson/meson8b.c > +++ b/drivers/clk/meson/meson8b.c > @@ -608,20 +608,27 @@ static struct clk_regmap meson8b_cpu_scale_div = { > }, > }; > > +static u32 mux_table_cpu_scale_out_sel[] = { 0, 1, 3 }; > static struct clk_regmap meson8b_cpu_scale_out_sel = { > .data = &(struct clk_regmap_mux_data){ > .offset = HHI_SYS_CPU_CLK_CNTL0, > .mask = 0x3, > .shift = 2, > + .table = mux_table_cpu_scale_out_sel, > }, > .hw.init = &(struct clk_init_data){ > .name = "cpu_scale_out_sel", > .ops = &clk_regmap_mux_ro_ops, > + /* > + * NOTE: We are skipping the parent with value 0x2 (which is > + * "cpu_div3") because it results in a duty cycle of 33% which > + * makes the system unstable and can result in a lockup of the > + * whole system. > + */ > .parent_names = (const char *[]) { "cpu_in_sel", > "cpu_div2", > - "cpu_div3", > "cpu_scale_div" }, > - .num_parents = 4, > + .num_parents = 3, > .flags = CLK_SET_RATE_PARENT, > }, > };
Reviewed-by: Jerome Brunet <jbrunet@baylibre.com>
| |