Hi Ilias,
On jeu., mars 14 2019, Ilias Apalodimas <ilias.apalodimas@xxxxxxxxxx> wrote:
> Hello Christian,
>> Hi,
>>
>> I assume you use the 1000 MHz firmware. This does also not work on my Rev 7
>> board. But I'm pretty sure this is not a problem of the patches, because if
>> I take a newer kernel (4.19.20/27) without the patches it also does not
>> work. A kernel 4.19.17 does work for me. My opinion on that is that this is
>> another problem which does just occure now because now the cpu frequency
>> scaling is working with the right frequencies.
> I am not sure which firmware i am running, i did all my tests on 5.0.0 and
> changing between governors worked fine without the patches
Curently my espressobin is broken so I tested the patches on the Armada
3700 DB and I didn't observe the issue you had.
The 3700 DB I used is configured to run at 800MHz.
Could you apply the following patch and sent me the boot log?
diff --git a/drivers/clk/mvebu/armada-37xx-periph.c b/drivers/clk/mvebu/armada-37xx-periph.c
index 26ed3c18a239..f814ade5cd80 100644
--- a/drivers/clk/mvebu/armada-37xx-periph.c
+++ b/drivers/clk/mvebu/armada-37xx-periph.c
@@ -452,14 +452,17 @@ static int clk_pm_cpu_set_parent(struct clk_hw *hw, u8 index)
/* Set the parent clock for all the load level */
for (load_level = 0; load_level < LOAD_LEVEL_NR; load_level++) {
- unsigned int reg, mask, val,
+ unsigned int reg, mask, val, old,
offset = ARMADA_37XX_NB_TBG_SEL_OFF;
armada_3700_pm_dvfs_update_regs(load_level, ®, &offset);
val = index << offset;
mask = ARMADA_37XX_NB_TBG_SEL_MASK << offset;
+ regmap_read(base, reg, &old);
regmap_update_bits(base, reg, mask, val);
+ pr_err("%s old=%X -> val=0x%X load_level=%d\n",
+ __func__, old, val, load_level);
}
return 0;
}
>
> Regards
> /Ilias
>>
>> Ilias Apalodimas <ilias.apalodimas@xxxxxxxxxx> schrieb am Do., 14. März
>> 2019, 13:15:
>>
>> > Hi Gregory,
>> > > The clock parenting was not setup properly when DVFS was enabled. It was
>> > > expected that the same clock source was used with and without DVFS which
>> > > was not the case.
>> > >
>> > > This patch fixes this issue, allowing to make the cpufreq support work
>> > > when the CPU clocks source are not the default ones.
>> > >
>> > > Fixes: 92ce45fb875d ("cpufreq: Add DVFS support for Armada 37xx")
>> > > Cc: <stable@xxxxxxxxxxxxxxx>
>> > > Reported-by: Christian Neubert <christian.neubert.86@xxxxxxxxx>
>> > > Reported-by: Ilias Apalodimas <ilias.apalodimas@xxxxxxxxxx>
>> > > Signed-off-by: Gregory CLEMENT <gregory.clement@xxxxxxxxxxx>
>> > > ---
>> > > drivers/clk/mvebu/armada-37xx-periph.c | 11 +++++++++++
>> > > 1 file changed, 11 insertions(+)
>> > >
>> > > diff --git a/drivers/clk/mvebu/armada-37xx-periph.c
>> > b/drivers/clk/mvebu/armada-37xx-periph.c
>> > > index 1f1cff428d78..26ed3c18a239 100644
>> > > --- a/drivers/clk/mvebu/armada-37xx-periph.c
>> > > +++ b/drivers/clk/mvebu/armada-37xx-periph.c
>> > > @@ -671,6 +671,17 @@ static int armada_3700_add_composite_clk(const
>> > struct clk_periph_data *data,
>> > > map = syscon_regmap_lookup_by_compatible(
>> > > "marvell,armada-3700-nb-pm");
>> > > pmcpu_clk->nb_pm_base = map;
>> > > +
>> > > + /*
>> > > + * Use the same parent when DVFS is enabled that the
>> > > + * default parent received at boot time. When this
>> > > + * function is called, DVFS is not enabled yet, so we
>> > > + * get the default parent and we can set the parent
>> > > + * for DVFS.
>> > > + */
>> > > + if (clk_pm_cpu_set_parent(muxrate_hw,
>> > > +
>> > clk_pm_cpu_get_parent(muxrate_hw)))
>> > > + dev_warn(dev, "Failed to setup default parent
>> > clock for DVFS\n");
>> > > }
>> > >
>> > > *hw = clk_hw_register_composite(dev, data->name,
>> > data->parent_names,
>> > > --
>> > > 2.20.1
>> > >
>> > Applied this and selected only
>> >
>> > CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y
>> > CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
>> > CONFIG_CPU_FREQ_GOV_POWERSAVE=y
>> >
>> > After changing the governor from 'powersave' to 'performance' the board
>> > completely froze (i even lost access to the serial port)
>> >
>> > Cheers
>> > /Ilias
>> >
--
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com
