Skip to contentSkip to conversation

phosphor-bittware repository

DISCUSSION [email protected] 2 participants 2 messages May 19, 2020

lore
Share & tools
lore
#0 lore raw

Hi Patrick

Quoted 54 lines

I looked briefly at the datasheet for this hardware [1]. It appears to

expose an SMBus interface for the features you mentioned. The most

straight-forward way to get this implemented is to create a kernel

driver for most of the features you mentioned. If you implement a

driver for this hardware that interacts with the hwmon, eeprom, and LED/GPIO

subsystems in the kernel(*), you'll be able to reuse a lot of existing

OpenBMC functionality without rewriting any userspace code.

- Sensor

- Kernel: hwmon

- Userspace: phosphor-hwmon or dbus-sensors

- VPD

- Kernel: eeprom

- Userspace: entity-manager (I think)

- LED control

- Kernel: LED / GPIO

- Userspace: phosphor-led-manager

The only part that wouldn't be covered would be the "Brick Protection".

We'd need to see some more information on how this is exposed but you

might be able to work it into the existing phosphor-bmc-code-mgmt

repository. For power supplies, they did create a separate repository

(phosphor-psu-code-mgmt) but I think they relied on some kernel APIs for

doing part of the work.

(*) Depending on how the bittware hardware is implemented at an SMBus

level you may end up with multiple smaller drivers (this is better).

Quoted 7 lines

If there is a single SMBus address for all these functions, you'll

likely end up with one big driver.

1. https://www.bittware.com/wp-content/uploads/datasheets/ds-250-soc.pdf

Sorry for late reply.

Because the 250-soc card needs to adjust the io expander to get the relevant information (e.g. temperature, VPD ...).

I think a dynamic detection function may be needed to handle the presence of the card and dynamically adjust the io expander.

On the other hand, I just want to be able to integrate all the functions.

Best Regards,
Ben

---------------------------------------------------------------------------------------------------------------------------------------------------------------
This email contains confidential or legally privileged information and is for the sole use of its intended recipient. Any unauthorized review, use, copying or distribution of this email or the content of this email is strictly prohibited. If you are not the intended recipient, you may reply to the sender and should delete this e-mail immediately.
---------------------------------------------------------------------------------------------------------------------------------------------------------------

#1 lore raw
Quoted 1 line · On Tue, May 19, 2020 at 06:47:48AM +0000, [email protected] wrote:

Because the 250-soc card needs to adjust the io expander to get the relevant information (e.g. temperature, VPD ...).

Is this IO expander on the i2c bus itself, like a mux or hub, or is it some specialized register set to switch between logic banks? In either case though, I don't know why you couldn't create a kernel driver for controlling access to the various logic blocks. If it is like an i2c-mux, there is a number of implementations of that in the kernel already.

Quoted 1 line

I think a dynamic detection function may be needed to handle the presence of the card and dynamically adjust the io expander.

I'm not sure what you mean by "dynamic detection function" here. If you just mean detecting the presence of the card by a GPIO or i2c probe call, entity-manager can handle some of that for you, I believe. You don't need to put the card into the device tree directly, but you could instead do some kind of probe call to ensure the device is present and then manually 'bind' the i2c address to a kernel driver. Entity-manager already supports doing this.

Quoted 1 line

On the other hand, I just want to be able to integrate all the functions.

I think you're going to end up duplicating a lot of code that already exists in the kernel this way and that is why I'm trying to steer you away from it. Just as a simple example on the sensors, we have two implementations that can already consume Linux-hwmon data and give you the dbus objects "for free". If you go the direction of putting it all in userspace, you're going to have to implement all the i2c activity, hwmon-like polling, and dbus objects all yourself. This isn't a trivial amount of code.

Signature · 1 line
Patrick Williams