[U-Boot] [PATCH 0/5] imx: ventana: stability fixes

[Tim Harvey]

On Wed, Jul 13, 2016 at 3:48 AM, Stefano Babic <sbabic at denx.de> wrote:
> Hi Tim,
>
> On 17/06/2016 15:01, Tim Harvey wrote:
>> This series addresses some stability issues related to voltage rails on
>> the IMX6 based Gateworks Ventana product family.
>>
>> Tim Harvey (5):
>>   imx: mx6: export the set_ldo_voltage function
>>   imx: ventana: fix PMIC rail adjustments for IMX6SDL in ldo-bypass mode
>>   pmic: ltc3676: add regulator config function
>>   imx: ventana: use continuous PWM mode and 1MHz for LTC3676 PMIC
>>     regulators
>>   imx: ventana: enable ldo-bypass mode on compatible kernels if
>>     ldobypass env set
>>
>>  arch/arm/cpu/armv7/mx6/soc.c              |   8 +-
>>  arch/arm/include/asm/arch-mx6/sys_proto.h |   9 ++
>>  board/gateworks/gw_ventana/common.c       | 109 +++++++++++++++------
>>  board/gateworks/gw_ventana/common.h       |   3 +-
>>  board/gateworks/gw_ventana/gw_ventana.c   | 152 +++++++++++++++++++++++++++++-
>>  drivers/power/pmic/pmic_ltc3676.c         |  16 ++++
>>  include/power/ltc3676_pmic.h              |  40 ++++++++
>>  7 files changed, 301 insertions(+), 36 deletions(-)
>>
>
> I do not find in my emailer 5/5 - I grab 0/5 for the answer.
>

Stefano,

If your missing 5/5 you can find it in my patches here:
https://patchwork.ozlabs.org/project/uboot/list/?submitter=41730

> Patch 4/5 is like a trick, but I am asking if the same solution can be
> used on other boards. The issue looks to me quite general, and if
> ldo-bypass must be set depends on if there is a capable PMIC.

4/5 'ventana: use continuous PWM mode and 1MHz for LTC3676 PMIC
regulators' was really about configuring the LTC3676 PMIC to use
'continuous PWM' instead of its power-on default of 'pulse skipping'
mode which was just noisy enough to cause issues with the IMX6 when it
was in ldo-bypass mode.

Is this the patch you were thinking of and what was it you felt could
be made more common?

There definitely is a 'trick' to determine if the kernel uses
ldo-bypass by examining its device-tree vddarm/vddsoc regulators in
'imx: ventana: fix PMIC rail adjustments for IMX6SDL in ldo-bypass
mode' and there is another 'trick' perhaps in 'imx: ventana: enable
ldo-bypass mode on compatible kernels if ldobypass env set' which
involves changing a device-tree that is found to 'support' LDO-bypass
mode but does not have it enabled but I'm not sure that is what you
meant.

>
> What about to let this code become more general moving into imx-common ?
> Do you see any issue why we shouldn't ?
>

I maybe should have split of this series into 3 sets but I figured I
would lump them all together as they were all 'power/PMIC' related:

1/5: imx: mx6: export the set_ldo_voltage function
2/5: imx: ventana: fix PMIC rail adjustments for IMX6SDL in ldo-bypass mode
  - these resolved a 1 to 5% board hang for ldo-bypass mode

3/5: pmic: ltc3676: add regulator config function
4/5: imx: ventana: use continuous PWM mode and 1MHz for LTC3676 PMIC regulators
  - these resolved a general board stability issue due to noise
sensitivity of the IMX6 when using ldo-bypass mode

5/5: imx: ventana: enable ldo-bypass mode on compatible kernels if
ldobypass env set
  - this will re-assign regulators in the device-tree to enable
ldo-bypass mode for device-tree's that are capable of running
ldo-bypass mode but have it disabled

As you likely know, IMX6 LDO-bypass requires board and kernel support
such that the kernel must have a PMIC driver capable of adjusting
vddarm and vddsoc within the range specified by the operation points
in the device-tree. Additionally, the device-tree must reference the
PMIC regulators and something in the kernel must put the IMX6 LDO's in
'bypass' mode by fully opening the FET's. Note that there is not yet
support for LDO-bypass in the mainline kernel, so this is strictly
used currently for Freescales downstream kernel currently.

I found that the recommended voltage ranges when in ldo-bypass mode vs
ldo-enabled mode don't overlap, specifically the maximum vddsoc for
imx6sdl in LDO-bypass mode exceeds the minimum for LDO-enabled mode. I
was able to prove that exceeding 1.21v for vddsoc when using
ldo-bypass mode does indeed cause an issue with the PCIe core (hangs
occasionally on init in the kernel). Therefore my only solution was to
determine which mode was going to be used by the booting kernel and
make adjustments while we could in the bootloader before booting the
kernel.

Now, after reviewing all of these to answer your questions I realize that:
 a) IMX6 ldo-bypass is not yet in the mainline Linux kernel
 b) IMX6 ldo-bypass may not be all that desirable anyway: the voltage
tolerances are much tighter which could cause stability issues, it
lowers the processor lifespan, and the power savings are not that
great in general use cases
 c) the two tricks mentioned above only resolve issues that occur on
boards with ldo-bypass enabled
 d) we switched our downstream Freescale vendor kernel a while back to
use ldo-enabled mode because we felt it was more stable and improved
part longevity. This was when I wrote '5/5: imx: ventana: enable
ldo-bypass mode on compatible kernels if ldobypass env set' to allow
users to still use it if their application was very power sensitive

So, I think I'll drop this series as it doesn't pertain to mainline
Linux kernels. I may re-write 4/5 and re-submit as I do feel that
continuous PWM mode for the LTC3676 PMIC may be more suitable for
sensitive rails in general.

Regards,

Tim